On Representations of Rational Cherednik Algebrasmath.mit.edu › ~sethsa › thesis.pdf · a...
Transcript of On Representations of Rational Cherednik Algebrasmath.mit.edu › ~sethsa › thesis.pdf · a...
On Representations of Rational Cherednik Algebras
by
Seth Shelley-Abrahamson
B.S., Stanford University (2013)
Submitted to the Department of Mathematicsin partial fulfillment of the requirements for the degree of
Doctor of Philosophy
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2018
c Massachusetts Institute of Technology 2018. All rights reserved.
Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Department of Mathematics
April 19, 2018
Certified by. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Pavel Etingof
Professor of MathematicsThesis Supervisor
Certified by. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Ivan Losev
Professor of MathematicsThesis Supervisor
Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .William Minicozzi
Co-Chair, Graduate Committee, Mathematics
2
On Representations of Rational Cherednik Algebras
by
Seth Shelley-Abrahamson
Submitted to the Department of Mathematicson April 19, 2018, in partial fulfillment of the
requirements for the degree ofDoctor of Philosophy
Abstract
This thesis introduces and studies two constructions related to the representation the-ory of rational Cherednik algebras: the refined filtration by supports for the categoryO and the Dunkl weight function.
The refined filtration by supports provides an analogue for rational Cherednikalgebras of the Harish-Chandra series appearing in the representation theory of fi-nite groups of Lie type. In particular, irreducible representations in the rationalCherednik category O with particular generalized support conditions correspond toirreducible representations of associated generalized Hecke algebras. An explicit pre-sentation for these generalized Hecke algebras is given in the Coxeter case, classifyingthe irreducible finite-dimensional representations in many new cases.
The Dunkl weight function K is a holomorphic family of tempered distributionson the real reflection representation of a finite Coxeter group W with values in linearendomorphisms of an irreducible representation of W. The distribution K gives rise toan integral formula for the Gaussian inner product on a Verma module in the rationalCherednik category O. At real parameter values, the restriction of K to the regularlocus in the real reflection representation can be interpreted as an analytic functiontaking values in Hermitian forms, invariant under the braid group, on the image ofa Verma module under the Knizhnik-Zamolodchikov (KZ) functor. This provides abridge between the study of invariant Hermitian forms on representations of rationalCherednik algebras and of Hecke algebras, allowing for a proof that the KZ functorpreserves signatures in an appropriate sense.
Thesis Supervisor: Pavel EtingofTitle: Professor of Mathematics
Thesis Supervisor: Ivan LosevTitle: Professor of Mathematics
3
4
AcknowledgmentsFirst, I would like to thank my advisors, Pavel Etingof and Ivan Losev, for theirpatience, kindness, and dedication. This thesis would not exist without them. Theyeach have taught me an enormous amount of mathematics and have generously sharedtheir ideas and insights with me throughout my Ph.D. These last several years at MIThave been some of the best in my life, and I owe both Pavel and Ivan an enormousamount for making my experience in graduate school so productive and enjoyable.Thank you.
Some acknowledgements relating to the mathematical content of this thesis arein order. The content of Chapter 3 of this thesis is closely based on my joint paper[52] with my co-advisor Ivan Losev. Many ideas appearing in that chapter, includingthe original idea for the project, are due to Ivan. I would also like to thank himfor countless useful conversations throughout the completion of that project, surelywithout which the project would never have been successful. That project furtherbenefited from Ivan’s conversations with Raphaël Rouquier, from my conversationswith Pavel Etingof and José Simental, and from comments from Emily Norton on apreliminary draft of [52].
Chapter 4 of this thesis is closely based on the preprint [67], which arose from aproject developed by Pavel Etingof and which benefited from his conversations withLarry Guth, Richard Melrose, Leonid Polterovich, and Vivek Shende. I would liketo express my deep gratitude to Pavel for suggesting that project, for many usefulconversations, for comments on a preliminary draft of [67], and for his ideas andinsights that permeate [67] and Chapter 4 of this thesis. I would also like to thankSemyon Dyatlov for providing essentially all of the content of Sections 4.4.1 and 4.4.2,notably including the crucial Lemma 4.4.2.2 and its proof, and for explaining to methe techniques from semiclassical analysis used in those sections.
Last, but certainly not least, I would like to thank my parents, Dana Shelley andPhil Abrahamson, and my fiancée, Olivia Hentz, for their constant love and support.I would be lost without you.
5
6
Contents
1 Introduction 11
2 Background and Definitions 152.1 Rational Cherednik Algebras . . . . . . . . . . . . . . . . . . . . . . . 152.2 PBW Theorem, Category 𝒪𝑐, and Supports . . . . . . . . . . . . . . 172.3 The Localization Lemma, Hecke Algebras, and the KZ Functor . . . . 212.4 Bezrukavnikov-Etingof Parabolic Induction and Restriction Functors . 23
2.4.1 Construction of Res𝑏 and Ind𝑏 . . . . . . . . . . . . . . . . . . 252.5 Partial KZ Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.6 Gordon-Martino Transitivity . . . . . . . . . . . . . . . . . . . . . . . 332.7 Mackey Formula for Rational Cherednik Algebras for Coxeter Groups 35
3 The Refined Filtration By Supports 393.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.2 Endomorphism Algebras Via Monodromy . . . . . . . . . . . . . . . . 47
3.2.1 Harish-Chandra Series for Rational Cherednik Algebras . . . . 473.2.2 The Functor 𝐾𝑍𝐿 . . . . . . . . . . . . . . . . . . . . . . . . . 523.2.3 Eigenvalues of Monodromy and Relations From Corank 1 . . . 59
3.3 The Coxeter Group Case . . . . . . . . . . . . . . . . . . . . . . . . . 663.3.1 A Presentation of the Endomorphism Algebra . . . . . . . . . 663.3.2 Computing Parameters Via KZ . . . . . . . . . . . . . . . . . 713.3.3 Type 𝐴 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.3.4 Type 𝐵 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863.3.5 Type 𝐷 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903.3.6 Parameters for Generalized Hecke Algebras in Exceptional Types 953.3.7 Type 𝐸 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 983.3.8 Type 𝐻 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1073.3.9 Type 𝐹4 with Unequal Parameters . . . . . . . . . . . . . . . 1093.3.10 Type 𝐼 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4 The Dunkl Weight Function 1234.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1234.2 Signature Characters and the Jantzen Filtration . . . . . . . . . . . . 126
4.2.1 Jantzen Filtrations on Standard Modules . . . . . . . . . . . . 1264.2.2 Hermitian Duals . . . . . . . . . . . . . . . . . . . . . . . . . 129
7
4.2.3 Characters and Signature Characters . . . . . . . . . . . . . . 1324.2.4 Rationality of Signature Characters . . . . . . . . . . . . . . . 1344.2.5 Asymptotic Signatures . . . . . . . . . . . . . . . . . . . . . . 1374.2.6 Isotypic Signature Characters . . . . . . . . . . . . . . . . . . 139
4.3 The Dunkl Weight Function . . . . . . . . . . . . . . . . . . . . . . . 1424.3.1 The Contravariant Pairing and Contravariant Form . . . . . . 1424.3.2 The Gaussian Pairing and Gaussian Inner Product . . . . . . 1444.3.3 Dunkl Weight Function: Characterization and Properties . . . 1464.3.4 Existence of Dunkl Weight Function for Small 𝑐 . . . . . . . . 1524.3.5 Analytic Continuation on the Regular Locus . . . . . . . . . . 1654.3.6 Extension to Tempered Distribution via the Wonderful Model 171
4.4 Signatures and the KZ Functor . . . . . . . . . . . . . . . . . . . . . 1784.4.1 Reminder on Semiclassical Analysis . . . . . . . . . . . . . . . 1794.4.2 A Key Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . 1824.4.3 Proof of the Signature Comparison Theorem . . . . . . . . . . 187
4.5 Conjectures and Further Directions . . . . . . . . . . . . . . . . . . . 1904.5.1 Complex Reflection Groups . . . . . . . . . . . . . . . . . . . 1904.5.2 Preservation of Jantzen Filtrations, Epsilon Factors, and Sig-
natures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1924.5.3 Local Description of 𝐾𝑐,𝜆 and Modules of Proper Support . . . 193
8
List of Tables
3.1 Parameters for Generalized Hecke Algebras . . . . . . . . . . . . . . . 973.1 (continued) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 983.2 Refined Filtration by Supports for 𝐸6 . . . . . . . . . . . . . . . . . . 1003.3 Refined Filtration by Supports for 𝐸7 . . . . . . . . . . . . . . . . . . 1023.4 Refined Filtration by Supports for 𝐸8 . . . . . . . . . . . . . . . . . . 1043.4 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1053.4 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1063.4 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1073.5 Refined Filtration by Supports for 𝐻3 . . . . . . . . . . . . . . . . . . 1073.5 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1083.6 Refined Filtration by Supports for 𝐻4 . . . . . . . . . . . . . . . . . . 1083.6 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1093.7 Simple Modules for 𝐻𝑐1,𝑐2(𝐹4) labeled by 𝐴′
1 . . . . . . . . . . . . . . 1113.8 Simple Modules for 𝐻𝑐1,𝑐2(𝐹4) labeled by 𝐴1 . . . . . . . . . . . . . . 1113.9 Simple Modules labeled by 𝐴′
1 × 𝐴1 . . . . . . . . . . . . . . . . . . . 1123.10 Simple Modules labeled by 𝐴′
2 . . . . . . . . . . . . . . . . . . . . . . 1123.11 Simple Modules labeled by 𝐴2 . . . . . . . . . . . . . . . . . . . . . . 1123.12 Simple Modules labeled by 𝐵2 . . . . . . . . . . . . . . . . . . . . . . 1123.12 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1133.13 Simple Modules labeled by 𝐵3 . . . . . . . . . . . . . . . . . . . . . . 1133.14 Simple Modules labeled by 𝐶3 . . . . . . . . . . . . . . . . . . . . . . 1133.15 Simple Modules labeled by 𝐴′
1 × 𝐴2 . . . . . . . . . . . . . . . . . . . 1133.16 Simple Modules labeled by 𝐴′
2 × 𝐴1 . . . . . . . . . . . . . . . . . . . 1143.17 Irreducible Representations of H𝑝,𝑞(𝐹4) . . . . . . . . . . . . . . . . . 1143.17 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1153.17 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1163.18 Modules of Given Support in 𝐻𝑐1,𝑐2(𝐹4) . . . . . . . . . . . . . . . . . 1173.18 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1183.18 continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
9
10
Chapter 1
Introduction
This thesis presents two new constructions in the representation theory of rational
Cherednik algebras. The first, the refined filtration by supports on the category
𝒪𝑐(𝑊, h) appearing in the paper [52], is discussed in Chapter 3 and permits the clas-
sification of the finite-dimensional irreducible representations of the rational Chered-
nik algebra 𝐻𝑐(𝑊, h) in many new cases. The second, the Dunkl weight function
introduced in the preprint [67], is discussed in Chapter 4 and provides an integral
formula for Cherednik’s Gaussian inner product, giving a bridge between the study
of invariant Hermitian forms on representations of rational Cherednik algebras and
Hecke algebras.
Let 𝑊 be a finite complex reflection group with reflection representation h, and let
𝑐 : 𝑆 → C be a 𝑊 -invariant complex-valued function on the set of complex reflections
𝑆 ⊂ 𝑊 . Associated to this data one has the rational Cherednik algebra 𝐻𝑐(𝑊, h),
introduced by Etingof and Ginzburg [29]. The algebras 𝐻𝑐(𝑊, h), parameterized by
such functions 𝑐 : 𝑆 → C, form a family of infinite-dimensional noncommutative asso-
ciative algebras providing a flat deformation of the algebra 𝐻0(𝑊, h) = C𝑊 n𝐷(h),
the semidirect product of 𝑊 with the algebra of polynomial differential operators on
h.
Since their introduction these algebras and their representation theory have been
intensely studied. Of particular interest is the category 𝒪𝑐(𝑊, h) of representations
of 𝐻𝑐(𝑊, h), introduced by Ginzburg, Guay, Opdam, and Rouquier [38], deforming
11
the category of finite-dimensional representations of 𝑊 and analogous to the clas-
sical Bernstein-Gelfand-Gelfand category 𝒪 attached to a finite-dimensional com-
plex semisimple Lie algebra g. In addition to being of interest from the purely
representation-theoretic point of view taken in this thesis, the category 𝒪𝑐(𝑊, h)
has proved to have connections with various other topics, including Hilbert schemes
[41, 42], torus knot invariants [30, 43], quantum integrable systems [25, 29, 34], and
categorification [65, 66].
The category 𝒪𝑐(𝑊, h) has many structures and properties in common with clas-
sical categories 𝒪. It is a highest weight category with standard objects labeled by
the set Irr(𝑊 ) of irreducible representations of the group 𝑊 ; to each 𝜆 ∈ Irr(𝑊 )
there is an associated standard module ∆𝑐(𝜆) with lowest weight 𝜆, analogous to the
Verma modules appearing in the representation theory of semisimple Lie algebras.
Each standard module ∆𝑐(𝜆) has a unique simple quotient 𝐿𝑐(𝜆), and the correspon-
dence 𝜆 ↦→ 𝐿𝑐(𝜆) provides a bijection between Irr(𝑊 ) and the isomorphism classes of
irreducible representations in 𝒪𝑐(𝑊, h). All of the finite-dimensional representations
of 𝐻𝑐(𝑊, h) belong to 𝒪𝑐(𝑊, h). Furthermore, each representation 𝑀 in 𝒪𝑐(𝑊, h) is
naturally graded and therefore has an associated character, defined to be the gener-
ating function recording the characters of the finite-dimensional representations of 𝑊
appearing in each graded component. However, unlike for semisimple Lie algebras,
explicit character formulas for irreducible representations and a classification of the
finite-dimensional representations remain unknown in many cases.
In the interest of keeping this thesis as self-contained as possible, and to fix no-
tations, Chapter 2 will provide relevant background and definitions. In particular,
Chapter 2 will recall definitions and basic results involving the rational Cherednik
algebra 𝐻𝑐(𝑊, h); its category 𝒪𝑐(𝑊, h) of representations and related constructions;
finite Hecke algebras H𝑞(𝑊 ) and the Knizhnik-Zamolodchikov (KZ) functor; the
Bezrukavnikov-Etingof parabolic induction and restriction functors; and the partial
KZ functors.
Chapter 3 introduces the refined filtration by supports for the category 𝒪𝑐(𝑊, h).
This is a filtration of 𝒪𝑐(𝑊, h) by Serre subcategories, refining the filtration by sup-
12
ports, with filtration subquotients labeled by 𝑊 -orbits of pairs (𝑊 ′, 𝐿) of a parabolic
subgroup 𝑊 ′ and an irreducible finite-dimensional representation 𝐿 of 𝐻𝑐(𝑊′, h𝑊 ′),
where h𝑊 ′ denotes the reflection representation of 𝑊 ′. The definition of this filtra-
tion is in direct analogy with the notion of Harish-Chandra series in the context
of representations of finite groups of Lie type, with the irreducible representations
appearing in the subquotient labeled by (𝑊 ′, 𝐿) viewed as belonging to the same
Harish-Chandra series. The subquotient category labeled by (𝑊 ′, 𝐿) is equivalent to
the category of finite-dimensional representations of the opposite endomorphism al-
gebra End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝, and the main technical result of Chapter 3 is the follow-
ing concrete description of these finite-dimensional algebras, appearing in Theorems
3.3.2.14 and 3.3.2.11:
Theorem. Let 𝑊 be a finite Coxeter group with simple reflections 𝑆, let 𝑐 : 𝑆 → C
be a class function, let 𝑊 ′ be a standard parabolic subgroup generated by the simple
reflections 𝑆 ′, and let 𝐿 be an irreducible finite-dimensional representation of the
rational Cherednik algebra 𝐻𝑐(𝑊′, h𝑊 ′). Let 𝑁𝑊 ′ denote the canonical complement
to 𝑊 ′ in its normalizer 𝑁𝑊 (𝑊 ′), let 𝑆𝑊 ′ ⊂ 𝑁𝑊 ′ denote the set of reflections in
𝑁𝑊 ′ with respect to its representation in the fixed space h𝑊′, and let 𝑁 𝑟𝑒𝑓
𝑊 ′ denote the
reflection subgroup of 𝑁𝑊 ′ generated by 𝑆𝑊 ′. Let 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ be a complement for 𝑁 𝑟𝑒𝑓
𝑊 ′ in
𝑁𝑊 ′, given as the stabilizer of a choice of fundamental Weyl chamber for the action
of 𝑁 𝑟𝑒𝑓𝑊 ′ on h𝑊
′. Then there is a class function 𝑞𝑊 ′,𝐿 : 𝑆𝑊 ′ → C× and an isomorphism
of algebras
End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝 ∼= 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ n H𝑞𝑊 ′,𝐿
(𝑁 𝑟𝑒𝑓𝑊 ′ )
where the semidirect product is defined by the action of 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ on H𝑞𝑊 ′,𝐿
(𝑁 𝑟𝑒𝑓𝑊 ′ ) by
diagram automorphisms arising from the conjugation action of 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ on 𝑁 𝑟𝑒𝑓
𝑊 ′ . Fur-
thermore, the parameter 𝑞𝑊 ′,𝐿 is explicitly computable.
The above theorem makes it possible to count, for any finite Coxeter group 𝑊 ,
the number of irreducible representations in 𝒪𝑐(𝑊, h) of any given support. In par-
ticular, this allows for the determination of the number of isomorphism classes of
finite-dimensional irreducible representations of 𝐻𝑐(𝑊, h). This number is computed
13
explicitly for all exceptional finite Coxeter groups 𝑊 , leading to the classification
of the finite-dimensional irreducible representations of 𝐻𝑐(𝑊, h) in many new cases.
Chapter 3 is closely based on the joint paper [52] with my co-advisor Ivan Losev.
Chapter 4 introduces the Dunkl weight function, generalizing previous results of
Dunkl to an arbitrary finite Coxeter group 𝑊 with an irreducible representation 𝜆.
The Dunkl weight function 𝐾𝑐,𝜆 is a family of EndC(𝜆)-valued tempered distributions,
holomorphic in the complex multi-parameter 𝑐, on the real reflection representation
of 𝑊 . For real parameters 𝑐, the distribution 𝐾𝑐,𝜆 provides an integral formula for
Cherednik’s Gaussian inner product 𝛾𝑐,𝜆 on the standard module ∆𝑐(𝜆):
Theorem. For any finite Coxeter group 𝑊 and irreducible representation 𝜆 of 𝑊 ,
there is a unique family 𝐾𝑐,𝜆, holomorphic in 𝑐, of EndC(𝜆)-valued tempered distri-
butions on the real reflection representation hR of 𝑊 such that the following integral
representation of the Gaussian pairing 𝛾𝑐,𝜆 holds:
𝛾𝑐,𝜆(𝑃,𝑄) =
∫hR
𝑄(𝑥)†𝐾𝑐,𝜆(𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥 for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆.
For all real parameters 𝑐, the restriction of the distribution 𝐾𝑐,𝜆 to the complement
hR,𝑟𝑒𝑔 of the reflection hyperplanes is given by integration against an analytic func-
tion that may be viewed as taking values in braid group-invariant Hermitian forms on
𝐾𝑍(∆𝑐(𝜆)). This provides a concrete connection between the study of invariant Her-
mitian forms on representations of rational Cherednik algebras and Hecke algebras,
permitting a proof that the KZ functor preserves signatures, and hence unitarizability,
in an appropriate sense. Chapter 4 is closely based on the preprint [67].
14
Chapter 2
Background and Definitions
In this chapter we will recall the definition of rational Cherednik algebras 𝐻𝑐(𝑊, h)
and some related objects and constructions that will be important later, including
the category of representations 𝒪𝑐(𝑊, h), standard modules ∆𝑐(𝜆), the KZ functor,
and the Bezrukavnikov-Etingof parabolic induction and restriction functors. Most of
the material recalled in this section can be found in [31], [38], and [5].
2.1 Rational Cherednik Algebras
A real reflection is an invertible linear transformation 𝑠 ∈ 𝐺𝐿(hR) of a finite dimen-
sional real vector space hR such that 𝑠2 = Id and dim ker(𝑠−1) = dim hR−1. A finite
real reflection group is a finite group 𝑊 along with a real faithful representation hR
such that the set of elements 𝑆 ⊂ 𝑊 acting in hR as real reflections generates 𝑊 .
We will refer to the representation hR as the real reflection representation of 𝑊 . A
finite group 𝑊 may be a finite real reflection group in more than one way, and we
will always consider the pair (𝑊, hR) of 𝑊 along with a given real reflection represen-
tation hR. The finite real reflection groups coincide with the finite Coxeter groups.
The standard classification of finite real reflection groups and their basic properties
can be found, for example, in [48].
Let h be a finite-dimensional complex vector space. A complex reflection is an
invertible linear operator 𝑠 ∈ 𝐺𝐿(h) of finite order such that rank(1 − 𝑠) = 1. Let
15
𝑊 ⊂ 𝐺𝐿(h) be a complex reflection group, i.e. a finite subgroup of 𝐺𝐿(h) generated
by the subset 𝑆 ⊂ 𝑊 of complex reflections lying in𝑊 . The complex reflection groups
were classified by Shephard and Todd in 1954 [69]; each such group decomposes as
a product of irreducible complex reflection groups, and every irreducible complex
reflection group either appears in the infinite family of complex reflection groups
𝐺(𝑚, 𝑝, 𝑛) indexed by integers 𝑚, 𝑝, 𝑛 ≥ 1 with 𝑝 | 𝑚 or is one of the 34 exceptional
irreducible complex reflection groups. Important special cases of complex reflection
groups include the finite real reflection groups, i.e. the finite Coxeter groups, which
may be regarded as complex reflection groups via complexification h = hR ⊗R C of
the real reflection representation hR. For example, the symmetric groups 𝑆𝑛 are the
complex reflection groups 𝐺(1, 1, 𝑛).
Fix a finite complex reflection group 𝑊 with complex reflection representation
h, and let 𝑆 ⊂ C denote the set of complex reflections in 𝑊 with respect to the
representation h. Let (·, ·) denote the natural pairing h* × h → C. Mimicking the
roots and coroots appearing in Lie theory, for each complex reflection 𝑠 ∈ 𝑆 choose
eigenvectors 𝛼𝑠 ∈ h* and 𝛼∨𝑠 ∈ h for 𝑠 with eigenvalues 𝜆𝑠 = 1 and 𝜆−1
𝑠 = 1, respec-
tively, partially normalized so that (𝛼𝑠, 𝛼∨𝑠 ) = 2. Let 𝑐 : 𝑆 → C be a 𝑊 -invariant
function with respect to the action of 𝑊 on 𝑆 by conjugation. We will refer to such
𝑐 as a parameter, and p will denote the C-vector space of such parameters. Given a
parameter 𝑐 ∈ p one may define a rational Cherednik algebra:
Definition 2.1.0.1. The rational Cherednik algebra 𝐻𝑐(𝑊, h) is the associative uni-
tal algebra with presentation
C𝑊 n 𝑇 (h⊕ h*)
⟨[𝑥, 𝑥′], [𝑦, 𝑦′], [𝑦, 𝑥] − (𝑦, 𝑥) +∑
𝑠∈𝑆 𝑐𝑠(𝛼𝑠, 𝑦)(𝑥, 𝛼∨𝑠 )𝑠 : 𝑥, 𝑥′ ∈ h*, 𝑦, 𝑦′ ∈ h⟩
.
In the definition above, 𝑇 (h⊕ h*) denotes the tensor algebra of h⊕ h*, and
C𝑊 n 𝑇 (h⊕ h*)
denotes the semidirect product algebra of 𝑇 (h ⊕ h*) with C𝑊 with respect to the
16
natural action of 𝑊 on 𝑇 (h⊕ h*): as a vector space we have
C𝑊 n 𝑇 (h⊕ h*) = C𝑊 ⊗ 𝑇 (h⊕ h*),
and the multiplication is given by (𝑤1⊗𝑡1)(𝑤2⊗𝑡2) = 𝑤1𝑤2⊗𝑤−12 (𝑡1)𝑡2. For example,
when 𝑐 = 0 the algebra 𝐻0(𝑊, h) is naturally identified with the algebra C𝑊 n𝐷(h),
where 𝐷(h) denotes the algebra of polynomial differential operators on h.
The rational Cherednik algebra 𝐻𝑐(𝑊, h) may alternatively be defined via its
faithful polynomial representation in the space C[h] in which an element 𝑤 ∈ 𝑊 acts
via the representation of 𝑊 on h, an element 𝑥 ∈ h* acts by multiplication, and an
element 𝑦 ∈ h acts by the Dunkl operator
𝐷𝑦 := 𝜕𝑦 −∑𝑠∈𝑆
2𝑐𝑠𝛼𝑠(𝑦)
(1 − 𝜆𝑠)𝛼𝑠(1 − 𝑠),
where 𝜕𝑦 denotes the derivative with respect to 𝑦.
2.2 PBW Theorem, Category 𝒪𝑐, and Supports
From the above presentation it is clear that 𝐻𝑐(𝑊, h) admits a filtration with deg𝑊 =
deg h = 0 and deg h* = 1, analogous to the usual filtration on 𝐷(h) by the or-
der of differential operators. Etingof and Ginzburg [29, Theorem 1.3] showed that
the associated graded algebra of 𝐻𝑐(𝑊, h) with respect to this filtration is naturally
identified with C𝑊 n 𝑆(h* ⊕ h), and in particular the natural multiplication map
C[h] ⊗ C𝑊 ⊗ C[h*] → 𝐻𝑐(𝑊, h) is an isomorphism of C-vector spaces:
Theorem 2.2.0.1 (PBW Theorem for Rational Cherednik Algebras). For any pa-
rameter 𝑐, the natural map
C[h] ⊗ C𝑊 ⊗ C[h*] → 𝐻𝑐(𝑊, h)
given by multiplication is an isomorphism of vector spaces.
17
This isomorphism should be viewed as an analogue for rational Cherednik algebras
of the triangular decomposition 𝑈(g) = 𝑈(n−)⊗𝑈(h)⊗𝑈(n+) of a complex semisimple
Lie algebra g with respect to a choice of Borel subalgebra b = h⊕ n+. The category
𝒪𝑐(𝑊, h) of representations of 𝐻𝑐(𝑊, h) introduced in [38] is then defined in parallel
with the classical Bernstein-Gelfand-Gelfand category 𝒪:
Definition 2.2.0.2 (Category 𝒪 for Rational Cherednik Algebras, [38]). The cate-
gory 𝒪𝑐(𝑊, h) is the full subcategory of the category of representations of 𝐻𝑐(𝑊, h)
consisting of those representations 𝑀 that are finitely generated over C[h] and on
which h ⊂ C[h*] acts locally nilpotently.
An important class of representations in 𝒪𝑐(𝑊, h) are the standard modules, which
are direct analogues of the classical Verma modules of semisimple Lie algebras:
Definition 2.2.0.3 (Standard Modules and Lowest Weight Representations). For
any irreducible representation 𝜆 ∈ Irr(𝑊 ), the standard module ∆𝑐(𝜆) ∈ 𝒪𝑐(𝑊, h)
with lowest weight 𝜆 is
∆𝑐(𝜆) := 𝐻𝑐(𝑊, h) ⊗C𝑊nC[h*] 𝜆,
where h ⊂ C[h*] acts on 𝜆 by 0. A module 𝑀 ∈ 𝒪𝑐(𝑊, h) will be called lowest weight
with lowest weight 𝜆 if 𝑀 is isomorphic to a nonzero quotient of ∆𝑐(𝜆).
Note that by the PBW theorem any standard module ∆𝑐(𝜆) is naturally isomor-
phic to C[h]⊗𝜆 as a module over C𝑊nC[h] ⊂ 𝐻𝑐(𝑊, h).We will use this identification
frequently.
Modules in the category 𝒪𝑐(𝑊, h) are naturally graded by their decomposition
into generalized eigenvectors for the grading element h ∈ 𝐻𝑐(𝑊, h), a deformation of
the Euler vector field:
Definition 2.2.0.4 (Grading Element). The grading element h ∈ 𝐻𝑐(𝑊, h) is the
element
h :=
dim h∑𝑖=1
𝑥𝑖𝑦𝑖 +dim h
2−∑𝑠∈𝑆
2𝑐𝑠1 − 𝜆𝑠
𝑠,
18
where 𝑥1, ..., 𝑥dim h is any basis of h* and 𝑦1, ..., 𝑦dim h is the associated dual basis of h.
Clearly, the element h does not depend on the choice of basis 𝑥1, ..., 𝑥dim h. When
𝑊 is a finite real reflection group, as will be the case most relevant for this thesis, we
have 𝜆𝑠 = −1 for all 𝑠 ∈ 𝑆, so h takes the form h =∑dim h
𝑖=1 𝑥𝑖𝑦𝑖 + dim h2
−∑
𝑠∈𝑆 𝑐𝑠𝑠.
From [31, Proposition 3.18, Theorem 3.28] we have:
Proposition 2.2.0.5. The element h ∈ 𝐻𝑐(𝑊, h) satisfies the commutation relations
[h, 𝑥] = 𝑥, 𝑥 ∈ h* [h, 𝑦] = −𝑦, 𝑦 ∈ h.
Furthermore, h acts locally finitely on any 𝑀 ∈ 𝒪𝑐(𝑊, h).
It follows that any 𝑀 ∈ 𝒪𝑐(𝑊, h) has a direct sum decomposition 𝑀 =⨁
𝑧∈C𝑀𝑧
into generalized eigenspaces for the action of h and that with respect to this decom-
position 𝑀 is a graded module with elements 𝑥 ∈ h* acting by degree 1 operators,
elements 𝑦 ∈ h acting by degree -1 operators, and elements 𝑤 ∈ 𝑊 acting by degree
0 operators.
This grading on ∆𝑐(𝜆) coincides with the usual grading on C[h] ⊗ 𝜆 shifted by
ℎ𝑐(𝜆), where ℎ𝑐(𝜆) = (dim h)/2 − 𝜒𝜆(∑
𝑠∈𝑆2𝑐𝑠1−𝜆𝑠 ) is the scalar by which the element
(dim h)/2−∑
𝑠∈𝑆2𝑐𝑠1−𝜆𝑠 𝑠 ∈ 𝑍(C𝑊 ) acts in the irreducible representation 𝜆 of𝑊 and 𝜒𝜆
denotes the character of the representation 𝜆. In particular, the h-weights appearing
nontrivially in the standard module are those in the set ℎ𝑐(𝜆) + 𝑛 : 𝑛 ∈ Z≥0. Any
proper submodule of ∆𝑐(𝜆) is graded and has all nontrivial h-weights appearing in the
set ℎ𝑐(𝜆) + 𝑛 : 𝑛 ∈ Z>0, and it follows that there is a maximal proper submodule
𝑁𝑐(𝜆) of ∆𝑐(𝜆). We denote the irreducible quotient ∆𝑐(𝜆)/𝑁𝑐(𝜆) by 𝐿𝑐(𝜆). By [31,
Proposition 3.30] we have
Proposition 2.2.0.6. Any irreducible representation 𝐿 ∈ 𝒪𝑐(𝑊, h) is isomorphic to
some 𝐿𝑐(𝜆) for a unique 𝜆 ∈ Irr(𝑊 ).
If 𝑁𝑐(𝜆) = 0, then there is a nonzero homomorphism ∆𝑐(𝜇) → 𝑁𝑐(𝜆) for some
𝜇 ∈ Irr(𝑊 ) such that ℎ𝑐(𝜇) = ℎ𝑐(𝜆) + 𝑘 for some integer 𝑘 > 0. The quantity
19
ℎ𝑐(𝜇) − ℎ𝑐(𝜆) is visibly a linear function of 𝑐, and 𝑐 ∈ p : ℎ𝑐(𝜇) = ℎ𝑐(𝜆) + 𝑘 is
an affine hyperplane in p. There are countably many such hyperplanes for various
𝜆, 𝜇 ∈ Irr(𝑊 ) and 𝑘 ∈ Z>0, and when 𝑐 lies on none of these hyperplanes it follows
that ∆𝑐(𝜆) = 𝐿𝑐(𝜆) for all 𝜆 ∈ Irr(𝑊 ). Furthermore, for such 𝑐 we have by a similar
argument that Ext1𝒪𝑐(𝑊,h)(∆𝑐(𝜆),∆𝑐(𝜇)) = 0 for all 𝜆, 𝜇 ∈ Irr(𝑊 ). In summary, we
have [31, Proposition 3.35]:
Proposition 2.2.0.7. For Weil generic 𝑐 ∈ p, the category 𝒪𝑐(𝑊, h) is semisimple
with simple objects 𝐿𝑐(𝜆) for 𝜆 ∈ Irr(𝑊 ).
Finally, we will need the notion of the support of a module 𝑀 ∈ 𝒪𝑐(𝑊, h):
Definition 2.2.0.8. Any 𝑀 ∈ 𝒪𝑐(𝑊, h) is by definition a finitely generated module
over C[h] with additional structure. Let Supp(𝑀) ⊂ h denote the support of 𝑀 .
When Supp(𝑀) = h we will say that 𝑀 has full support.
For example, the standard module ∆𝑐(𝜆) is a finitely generated free module over C[h],
and in particular Supp(∆𝑐(𝜆)) = h.
The determination of the support varieties Supp(𝐿𝑐(𝜆)) of the irreducible repre-
sentations 𝐿𝑐(𝜆) is a fundamental question about 𝒪𝑐(𝑊, h). Using parabolic induction
and restriction functors, Bezrukavnikov and Etingof [5] showed that the subvarieties
of h appearing as supports of irreducible representations in 𝒪𝑐(𝑊, h) can only be the
form 𝑋(𝑊 ′) := 𝑊h𝑊′ where 𝑊 ′ is a parabolic subgroup of 𝑊 , i.e. the stabilizer of a
point 𝑏 ∈ h. Note that 𝑋(𝑊 ′) = 𝑋(𝑊 ′′) exactly when 𝑊 ′ and 𝑊 ′′ are conjugate in
𝑊 , and in particular the possible supports of simple modules in 𝒪𝑐(𝑊, h) are indexed
by conjugacy classes of parabolic subgroups of 𝑊 . It is important to note that not
necessarily all such varietes 𝑋(𝑊 ′) appear as the support varieties of representations
in 𝒪𝑐(𝑊, h), and in fact for generic values of the parameter 𝑐 all irreducible repre-
sentations in 𝒪𝑐(𝑊, h) have full support, in which case only the variety 𝑋(1) = h
appears.
20
2.3 The Localization Lemma, Hecke Algebras, and
the KZ Functor
Let h𝑟𝑒𝑔 := h∖⋃𝑠∈𝑆 ker(𝛼𝑠), the regular locus for the action of 𝑊 on h, be the com-
plement in h of the arrangement of reflection hyperplanes for the action of 𝑊 on h.
Note that h𝑟𝑒𝑔 is precisely the principal affine open subset of h defined by the non-
vanishing of the discriminant element 𝛿 :=∏
𝑠∈𝑆 𝛼𝑠. As 𝛿|𝑊 | is 𝑊 -invariant and has
locally nilpotent adjoint action on 𝐻𝑐(𝑊, h), it follows that the noncommutative lo-
calization 𝐻𝑐(𝑊, h)[𝛿−1] coincides with the localization of 𝐻𝑐(𝑊, h) as a C[h]-module,
and this localized algebra is isomorphic to the algebra C𝑊 n 𝐷(h𝑟𝑒𝑔) in a natural
way (see [38, Theorem 5.6]). Thus, a module 𝑀 in 𝒪𝑐(𝑊, h) determines a module
𝑀 [𝛿−1] over C𝑊n𝐷(h𝑟𝑒𝑔) by localization, which may be regarded as a 𝑊 -equivariant
𝐷-module on h𝑟𝑒𝑔. The 𝑊 -equivariant 𝐷(h𝑟𝑒𝑔)-modules occurring in this way are
𝒪(h𝑟𝑒𝑔)-coherent, by definition of the category 𝒪𝑐(𝑊, h), and have regular singular-
ities [38]. By the Riemann-Hilbert correspondence [16], descending to h𝑟𝑒𝑔/𝑊 and
taking the monodromy representation at a base point 𝑏 ∈ h𝑟𝑒𝑔 defines an equivalence
of categories
C𝑊 n𝐷(h𝑟𝑒𝑔)-mod𝒪(h𝑟𝑒𝑔)-coh, r.s.∼= 𝜋1(h𝑟𝑒𝑔/𝑊, 𝑏)-mod𝑓.𝑑.
where the subscript 𝒪(h𝑟𝑒𝑔)-coh, r.s. indicates that the 𝐷(h𝑟𝑒𝑔)-modules are 𝒪(h𝑟𝑒𝑔)-
coherent with regular singularities and the subscript 𝑓.𝑑. indicates that 𝜋1(h𝑟𝑒𝑔/𝑊, 𝑏)-
modules are finite-dimensional. This procedure of localization to h𝑟𝑒𝑔 followed by
descent to h𝑟𝑒𝑔 and the Riemann-Hilbert correspondence thus defines an exact functor
𝒪𝑐(𝑊, h) → 𝜋1(h𝑟𝑒𝑔/𝑊, 𝑏)-mod𝑓.𝑑..
The fundamental group 𝜋1(h𝑟𝑒𝑔/𝑊, 𝑏) is the generalized braid group attached to
𝑊 and is denoted 𝐵𝑊 . In [38], it was shown that the above functor in fact factors
through the category H𝑞(𝑊 )-mod𝑓.𝑑. of finite-dimensional representations over a cer-
21
tain quotient H𝑞(𝑊 ), the Hecke algebra, of the group algebra C𝐵𝑊 . The resulting
exact functor
𝐾𝑍 : 𝒪𝑐(𝑊, h) → H𝑞(𝑊 )-mod𝑓.𝑑.
is the 𝐾𝑍 functor. The quotient H𝑞(𝑊 ) is defined as follows [7]. Let ℋ = ker(𝑠) :
𝑠 ∈ 𝑆 be the set of reflection hyperplanes for the action of 𝑊 on h. For each
reflection hyperplane 𝐻 ∈ ℋ, let 𝑇𝐻 be a representative of the conjugacy class in
𝐵𝑊 determined by a small loop, oriented counterclockwise, in h𝑟𝑒𝑔/𝑊 about the
hyperplane 𝐻. The group 𝐵𝑊 is generated by the union of these conjugacy classes
([7, Theorem 2.17]). For each 𝐻 ∈ ℋ, let 𝑙𝐻 be the order of the cyclic subgroup of
𝑊 stabilizing 𝐻 pointwise, and let 𝑞1,𝐻 , ...𝑞𝑙𝐻−1,𝐻 ∈ C× be nonzero complex numbers
that are 𝑊 -invariant, i.e. 𝑞𝑗,𝐻 = 𝑞𝑗,𝐻′ whenever 𝐻 = 𝑤𝐻 ′ for some 𝑤 ∈ 𝑊 . Let 𝑞
denote the collection of these 𝑞𝑗,𝐻 , and define the Hecke algebra H𝑞(𝑊 ) as the quotient
of C𝐵𝑊 by the relations
(𝑇𝐻 − 1)
𝑙𝐻−1∏𝑗=1
(𝑇𝐻 − 𝑒2𝜋𝑖𝑗/𝑙𝐻𝑞𝑗,𝐻) = 0.
The choice of the parameter 𝑞 such that the functor 𝐾𝑍 is defined as above is given
explicitly in [38] as a function of the parameter 𝑐 for the algebra 𝐻𝑐(𝑊, h). In the
special case that 𝑊 is a real reflection group, which is the primary case relevant to
this thesis, the dependence of 𝑞 on 𝑐 is especially simple. In that case we have 𝑙𝐻 = 2
for all 𝐻 ∈ ℋ, and 𝑞1,𝐻 = 𝑒−2𝜋𝑖𝑐𝑠 where 𝑠 ∈ 𝑆 is the reflection such that 𝐻 = ker(𝑠)
(note that the sign convention for 𝑐 in this thesis differs from that appearing in
[38]). In particular, in the Coxeter case the Hecke algebra H𝑞(𝑊 ) appearing in the
𝐾𝑍 functor is precisely the Iwahori-Hecke algebra attached to the Coxeter group 𝑊
with generators 𝑇𝑠 : 𝑠 ∈ 𝑆 satisfying the relations in the braid group 𝐵𝑊 and the
quadratic relations
(𝑇𝑠 − 1)(𝑇𝑠 + 𝑒−2𝜋𝑖𝑐𝑠) = 0.
When it is relevant, we will explicitly emphasize the dependence of the functor
𝐾𝑍 on the base point 𝑏 ∈ h𝑟𝑒𝑔 using the notation 𝐾𝑍𝑏. Clearly, the functor 𝐾𝑍𝑏 does
22
not depend on 𝑏 ∈ h𝑟𝑒𝑔 up to isomorphism. At the level of vector spaces, 𝐾𝑍𝑏(𝑀) is
the fiber at 𝑏 of the C[h]-module 𝑀 ∈ 𝒪𝑐(𝑊, h). As a consequence, 𝐾𝑍𝑏(𝑀) = 0 if
and only if 𝑀 has full support.
In Chapter 4, we will be particularly concerned with the image of the standard
modules ∆𝑐(𝜆) under 𝐾𝑍𝑏, and it is worth considering the 𝐾𝑍𝑏(∆𝑐(𝜆)) in some
detail here. Recall that by the PBW theorem for rational Cherednik algebras we
have ∆𝑐(𝜆) = C[h] ⊗ 𝜆 as a C𝑊 n C[h]-module, giving rise to an identification
∆𝑐(𝜆)[𝛿−1] = C[h𝑟𝑒𝑔] ⊗ 𝜆 of C𝑊 n C[h𝑟𝑒𝑔]-modules. For any 𝑦 ∈ h ⊂ 𝐻𝑐(𝑊, h) we
have 𝑦𝜆 = 0, by the definition of ∆𝑐(𝜆). At the same time, with respect to the natural
identification of 𝐻𝑐(𝑊, h)[𝛿−1] with C𝑊 n𝐷(h𝑟𝑒𝑔) the element 𝑦 ∈ h corresponds to
the Dunkl operator 𝐷𝑦 = 𝜕𝑦 −∑
𝑠∈𝑆 𝑐𝑠𝛼𝑠(𝑦)𝛼𝑠
(1 − 𝑠). In particular, it follows that the
vector field 𝜕𝑦 ∈ C𝑊 n𝐷(h𝑟𝑒𝑔) acts on 𝜆 ⊂ ∆𝑐(𝜆)[𝛿−1] by
𝜕𝑦𝑣 =∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(1 − 𝑠)
for all 𝑣 ∈ 𝜆 ⊂ ∆𝑐(𝜆)[𝛿−1]. It folows that the 𝑊 -equivariant 𝐷(h𝑟𝑒𝑔)-module structure
on ∆𝑐(𝜆)[𝛿−1] arises from the flat KZ connection
∇𝐾𝑍 := 𝑑+∑𝑠∈𝑆
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
(1 − 𝑠)
on the trivial vector bundle on h𝑟𝑒𝑔 with fiber 𝜆.
2.4 Bezrukavnikov-Etingof Parabolic Induction and
Restriction Functors
Parabolic induction and restriction functors are central to the study of representations
of Iwahori-Hecke algebras. The definition of these functors relies on the natural
embedding H𝑞(𝑊′) ⊂ H𝑞(𝑊 ) of the Hecke algebra H𝑞(𝑊
′) attached to a parabolic
subgroup 𝑊 ′ of a Coxeter group 𝑊 in the Hecke algebra H𝑞(𝑊 ) attached to 𝑊 .
Parabolic restriction is then simply the naive restriction, and parabolic induction is
23
the tensor product H𝑞(𝑊 ) ⊗H𝑞(𝑊 ′) ∙. As for Hecke algebras, parabolic induction and
restriction functors are central to the study of representations of rational Cherednik
algebras 𝐻𝑐(𝑊, h). Let h𝑊 ′ denote the unique 𝑊 ′-stable complement to h𝑊′ in h.
Then 𝑊 ′ acts on h𝑊 ′ , and the action is generated by reflections. Let 𝐻𝑐(𝑊′, h𝑊 ′)
denote the associated rational Cherednik algebra, where by abuse of notation 𝑐 here
denotes the restriction of 𝑐 to the reflections 𝑆 ′ = 𝑆 ∩𝑊 ′ in 𝑊 lying in 𝑊 ′. Then,
unlike for Hecke algebras, the algebra 𝐻𝑐(𝑊′, h𝑊 ′) does not embed as a subalgebra of
𝐻𝑐(𝑊, h) in a natural way, and the restriction and induction functors must be defined
differently.
Bezrukavnikov and Etingof [5] circumvented this difficulty by using the geometric
interpretation of rational Cherednik algebras. The price to pay for this approach is
that rather than defining a single parabolic restriction functor
Res𝑊𝑊 ′ : 𝒪𝑐(𝑊, h) → 𝒪𝑐(𝑊′, h𝑊 ′)
and a single parabolic induction functor
Ind𝑊𝑊 ′ : 𝒪𝑐(𝑊′, h𝑊 ′) → 𝒪𝑐(𝑊, h),
one instead defines a restriction functor Res𝑏 and induction functor Ind𝑏 defined for
every point 𝑏 in the locally closed subvariety h𝑊′
𝑟𝑒𝑔 of h consisting of those points 𝑏 ∈ h
with stabilizer Stab𝑊 (𝑏) = 𝑊 ′. Bezrukavnikov and Etingof explain the dependence
of Res𝑏 on the choice of 𝑏 ∈ h𝑊′
𝑟𝑒𝑔 in the following elegant manner. They construct an
exact functor
Res𝑊𝑊 ′ : 𝒪𝑐(𝑊, h) → (𝒪𝑐(𝑊′, h𝑊 ′) Loc(h𝑊
′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′),
where Loc(h𝑊′
𝑟𝑒𝑔) denotes the category of local systems on h𝑊′
𝑟𝑒𝑔 and the 𝑁𝑊 (𝑊 ′) in the
superscript indicates equivariance for the normalizer 𝑁𝑊 (𝑊 ′) of 𝑊 ′ in 𝑊 , and they
show that the fiber (Res𝑊𝑊 ′)𝑏 of the resulting local system at 𝑏 ∈ h𝑊′
𝑟𝑒𝑔 gives precisely
the parabolic restriction functor Res𝑏. In particular, for any points 𝑏, 𝑏′ ∈ h𝑊′
𝑟𝑒𝑔, parallel
24
transport along any path 𝛾 in h𝑊′
𝑟𝑒𝑔 connecting 𝑏 and 𝑏′ defines an isomorphism between
the functors Res𝑏 and Res𝑏′ . As explained in [5], in the case 𝑊 ′ = 1, the functor Res𝑊𝑊 ′
can be identified with the 𝐾𝑍 functor recalled above in Section 2.3, and therefore the
functors Res𝑊𝑊 ′ can be regarded as relative versions of the 𝐾𝑍 functor. We refer to
the functors Res𝑊𝑊 ′ as the partial KZ functors, and these functors are central to the
approach we take here.
We recall the construction of the functors Res𝑏, Ind𝑏, and Res𝑊𝑊 ′ below, following
[5, Section 3].
2.4.1 Construction of Res𝑏 and Ind𝑏
Let 𝑏 ∈ h𝑊′
𝑟𝑒𝑔, i.e. a point in h with stabilizer 𝑊 ′ in 𝑊 . The completion 𝐻𝑐(𝑊, h)𝑏
of the rational Cherednik algebra 𝐻𝑐(𝑊, h) at the orbit 𝑊𝑏 ⊂ h is defined to be the
associative algebra 𝐻𝑐(𝑊, h)𝑏 := C[h]∧𝑊𝑏 ⊗C[h] 𝐻𝑐(𝑊, h),
where C[h]∧𝑊𝑏 denotes the completion of C[h] at the orbit 𝑊𝑏. Restricting the pa-
rameter 𝑐 : 𝑆 → C to the set of reflections 𝑆 ′ ⊂ 𝑊 ′, one may form the rational
Cherednik algebra 𝐻𝑐(𝑊′, h) and its completion 𝐻𝑐(𝑊
′, h)0 at 0 ∈ h. As explained
in [5, Section 3.3], one may think of the algebra 𝐻𝑐(𝑊, h)𝑏 in a more geometric man-
ner as the subalgebra of EndC(C[h]∧𝑊𝑏) generated by C[h]∧𝑊𝑏 , the group 𝑊 , and the
Dunkl operators 𝐷𝑦 associated to points 𝑦 ∈ h. This geometric interpretation leads
naturally to an isomorphism ([5, Theorem 3.2])
𝜃 : 𝐻𝑐(𝑊, h)𝑏 → 𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊′, h)0)
where the algebra 𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊′, h)0), the centralizer algebra, is the endomorphism
algebra of the right 𝐻𝑐(𝑊, h)0-module 𝐹𝑢𝑛𝑊 ′(𝑊, 𝐻𝑐(𝑊′, h)0) of functions 𝑓 : 𝑊 →𝐻𝑐(𝑊
′, h)0 satisfying 𝑓(𝑤′𝑤) = 𝑤′𝑓(𝑤) for all 𝑤 ∈ 𝑊,𝑤′ ∈ 𝑊 ′. We may take the
completion 𝐻𝑐(𝑊′, h)0 of 𝐻𝑐(𝑊
′, h) at 0 rather than at 𝑏 in the centralizer because
the assignments 𝑤′ ↦→ 𝑤′ for 𝑤′ ∈ 𝑊 ′, 𝑦 ↦→ 𝑦 for 𝑦 ∈ h, and 𝑥 ↦→ 𝑥− 𝑏 extends to an
25
isomorphism 𝐻𝑐(𝑊′, h)0 ∼= 𝐻𝑐(𝑊
′, h)𝑏. The isomorphism 𝜃 determines, by transfer of
structure, an equivalence of categories
𝜃* : 𝐻𝑐(𝑊, h)𝑏-mod → 𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊′, h)0)-mod.
The centralizer algebra 𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊′, h)0) is non-canonically isomorphic to the
matrix algebra of size |𝑊/𝑊 ′| over 𝐻𝑐(𝑊′, h)0, and in particular the isomorphism 𝜃
shows that 𝐻𝑐(𝑊, h)𝑏 is Morita equivalent to 𝐻𝑐(𝑊′, h)0. A particularly natural choice
of Morita equivalence is given by the idempotent 𝑒𝑊 ′ ∈ 𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊′, h)0) defined
by 𝑒𝑊 ′(𝑓)(𝑤) = 𝑓(𝑤) for 𝑤 ∈ 𝑊 ′ and 𝑒𝑊 ′(𝑓)(𝑤) = 0 for 𝑤 /∈ 𝑊 ′. Note that 𝑒𝑊 ′ is
the image under 𝜃* of the idempotent 1𝑏 ∈ C[h]∧𝑊𝑏 ⊂ 𝐻𝑐(𝑊, h)𝑏 that is the indicator
function of the point 𝑏 in its 𝑊 -orbit. As the two-sided ideal in the centralizer algebra
generated by 𝑒𝑊 ′ is the entire centralizer algebra and as the associated spherical
subalgebra 𝑒𝑊 ′𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊′, h)0)𝑒𝑊 ′ is naturally identified with 𝐻𝑐(𝑊
′, h)0, the
functors
𝐼 : 𝐻𝑐(𝑊′, h)0-mod → 𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊
′, h)0)-mod
𝑀 ↦→ 𝐼(𝑀) := 𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊′, h)0)𝑒𝑊 ′ ⊗𝑒𝑊 ′𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊 ′,h)0)𝑒𝑊 ′
𝑀
and
𝐼−1 : 𝑍(𝑊,𝑊 ′, 𝐻𝑐(𝑊′, h)0)-mod → 𝐻𝑐(𝑊
′, h)0-mod
𝑁 ↦→ 𝐼−1(𝑁) := 𝑒𝑊 ′𝑁
are mutually quasi-inverse equivalences.
Let 𝒪𝑐(𝑊, h)𝑏 denote the full subcategory of 𝐻𝑐(𝑊, h)𝑏-modules which are finitely
generated over C[h]∧𝑊𝑏 . Let
𝑏 : 𝒪𝑐(𝑊, h) → 𝒪𝑐(𝑊, h)𝑏
be the functor of completion at the orbit 𝑊𝑏 and let
𝐸𝑏 : 𝒪𝑐(𝑊, h)𝑏 → 𝒪𝑐(𝑊, h)
26
be the functor that sends a module 𝑀 to its subspace 𝐸𝑏(𝑀) of h-locally nilpotent
vectors. By [5, Proposition 3.6, Proposition 3.8], these functors are exact and 𝐸𝑏 is
the right adjoint of 𝑏. Similarly, we have the category 𝒪𝑐(𝑊′, h)0, the completion
functor 0 : 𝒪𝑐(𝑊′, h) → 𝒪𝑐(𝑊
′, h)0 and the functor 𝐸0 : 𝒪𝑐(𝑊′, h)0 → 𝒪𝑐(𝑊
′, h)
taking h-locally nilpotent vectors, which are in fact category equivalences [5, Theorem
2.3].
Let h𝑊 ′ denote the unique 𝑊 ′-stable complement to h𝑊′ in h. Clearly, an element
𝑠 ∈ 𝑊 ′ acts on h as a complex reflection if and only if it acts on h𝑊 ′ as a complex
reflection, and 𝑊 ′ acts on h𝑊 ′ faithfully. By restriction of the parameter 𝑐 : 𝑆 →
C to the set of reflections 𝑆 ′ in 𝑊 ′, we may form the rational Cherednik algebra
𝐻𝑐(𝑊′, h𝑊 ′) and its associated category 𝒪𝑐(𝑊
′, h𝑊 ′). Let
𝜁 : 𝒪𝑐(𝑊′, h) → 𝒪𝑐(𝑊
′, h𝑊 ′)
be the functor defined by
𝜁(𝑀) = 𝑚 ∈𝑀 : 𝑦𝑚 = 0 ∀𝑦 ∈ h𝑊′.
As discussed in [5, Section 2.3], 𝜁 is an equivalence of categories - this is essentially
an instance of Kashiwara’s lemma for 𝐷(h𝑊′)-modules, in view of the natural tensor
product decomposition 𝐻𝑐(𝑊′, h) ∼= 𝐻𝑐(𝑊
′, h𝑊 ′) ⊗𝐷(h𝑊′).
Definition 2.4.1.1. The parabolic restriction functor Res𝑏 : 𝒪𝑐(𝑊, h) → 𝒪𝑐(𝑊′, h𝑊 ′)
is the composition
Res𝑏 := 𝜁 ∘ 𝐸0 ∘ 𝐼−1 ∘ 𝜃* ∘ 𝑏and the parabolic induction functor Ind𝑏 : 𝒪𝑐(𝑊
′, h𝑊 ′) → 𝒪𝑐(𝑊, h) is the composition
𝐸𝑏 ∘ 𝜃−1* ∘ 𝐼 ∘ 0 ∘ 𝜁−1.
Proposition 2.4.1.2. The functors Ind𝑏 and Res𝑏 are exact and biadjoint and do not
depend on the choice of 𝑏 ∈ h𝑊′
𝑟𝑒𝑔 up to isomorphism.
27
Proof. Exactness is [5, Proposition 3.9(i)], and that Ind𝑏 is right adjoint to Res𝑏 is
[5, Theorem 3.10]. That Ind𝑏 is also left adjoint to Res𝑏 was established by Shan [65,
Section 2.4] under some assumptions and later in full generality by Losev [54]. The
independence up to isomorphism on the choice of 𝑏 was established in [5, Section 3.7]
using the partial 𝐾𝑍 functor Res𝑊𝑊 ′ to be recalled in the following section.
2.5 Partial KZ Functors
In this section we give in some detail a construction of the partial KZ functor Res𝑊𝑊 ′
introduced in [5, Section 3.7]. Let 𝐻𝑐(𝑊, h)𝑊h𝑊′
𝑟𝑒𝑔denote the completion of 𝐻𝑐(𝑊, h) at
the 𝑊 -stable locally closed subvariety 𝑊h𝑊′
𝑟𝑒𝑔 ⊂ h. As for the case of completion at the
𝑊 -orbit of a point, the completion 𝐻𝑐(𝑊, h)𝑊h𝑊′
𝑟𝑒𝑔can be realized as the subalgebra of
EndC(C[h]∧𝑊h𝑊
′𝑟𝑒𝑔 ) generated by the functions C[h]
∧𝑊h𝑊
′𝑟𝑒𝑔 on the formal neighborhood
of 𝑊h𝑊′
𝑟𝑒𝑔 in h, the group 𝑊 , and the Dunkl operators 𝐷𝑦 for 𝑦 ∈ h associated to
the action of 𝑊 on h. (Recall that the algebra of functions C[h]∧𝑊h𝑊
′𝑟𝑒𝑔 is obtained
by first localizing to a principal open subset in which 𝑊h𝑊′
𝑟𝑒𝑔 is a closed subvariety,
followed by completion at the ideal defining 𝑊h𝑊′
𝑟𝑒𝑔 in this principal open subset.)
Note that the variety 𝑊h𝑊′
𝑟𝑒𝑔 is the disjoint union of the 𝑊 -translates of h𝑊 ′𝑟𝑒𝑔, and the
set-wise stabilizer of h𝑊 ′𝑟𝑒𝑔 in 𝑊 is precisely 𝑁𝑊 (𝑊 ′). Let 1h𝑊
′𝑟𝑒𝑔
∈ C[h]∧𝑊h𝑊
′𝑟𝑒𝑔 denote
the indicator function of h𝑊 ′𝑟𝑒𝑔 as a function on the formal neighborhood in h of its
𝑊 -orbit. Similar to the isomorphism
1𝑏 𝐻𝑐(𝑊, h)𝑏1𝑏 ∼= 𝐻𝑐(𝑊′, h)𝑏
from [5] discussed in the previous section, we have a natural isomorphism of algebras
1h𝑊′
𝑟𝑒𝑔
𝐻𝑐(𝑊, h)𝑊h𝑊′
𝑟𝑒𝑔1h𝑊
′𝑟𝑒𝑔
∼= 𝐻𝑐(𝑁𝑊 (𝑊 ′), h)h𝑊 ′𝑟𝑒𝑔,
where the algebra 𝐻𝑐(𝑁𝑊 (𝑊 ′), h)h𝑊 ′𝑟𝑒𝑔
denotes the completion of the algebra
𝐻𝑐(𝑁𝑊 (𝑊 ′), h) = C𝑁𝑊 (𝑊 ′) n𝑊 ′ 𝐻𝑐(𝑊′, h)
28
at the 𝑁𝑊 (𝑊 ′)-stable locally closed subvariety h𝑊′
𝑟𝑒𝑔 ⊂ h (for comparison with the
case of étale pullbacks rather than completions, see [45, Theorem 3.2]). The formal
neighborhood of h𝑊 ′𝑟𝑒𝑔 inside h is canonically identified with its formal neighborhood
inside the principal open subset h𝑊 ′𝑟𝑒𝑔 × h𝑊 ′ ⊂ h, which is precisely (h𝑊
′𝑟𝑒𝑔 × h𝑊 ′)
∧h𝑊
′𝑟𝑒𝑔 =
h𝑊′
𝑟𝑒𝑔×h∧0
𝑊 ′ and has ring of formal functions C[h𝑊′]⊗C[[h𝑊 ′ ]]. From this it follows im-
mediately that there is a natural isomorphism
𝐻𝑐(𝑁𝑊 (𝑊 ′), h)h𝑊 ′𝑟𝑒𝑔
∼= 𝑁𝑊 (𝑊 ′) n𝑊 ′ ( 𝐻𝑐(𝑊′, h𝑊 ′)0⊗𝐷(h𝑊
′
𝑟𝑒𝑔)).
Let
𝜓 : 1h𝑊′
𝑟𝑒𝑔
𝐻𝑐(𝑊, h)𝑊h𝑊′
𝑟𝑒𝑔1h𝑊
′𝑟𝑒𝑔
→ 𝑁𝑊 (𝑊 ′) n𝑊 ′ ( 𝐻𝑐(𝑊′, h𝑊 ′)0⊗𝐷(h𝑊
′
𝑟𝑒𝑔))
be the isomorphism obtained as the composition of the two natural isomorphisms
discussed above, and let 𝜓* denote the induced equivalence of module categories.
As in Definition 2.2.0.4, let h𝑊 ′ ∈ 𝐻𝑐(𝑊′, h𝑊 ′) denote the grading element
h𝑊 ′ :=𝑛∑𝑖=1
𝑥𝑖𝑦𝑖 +𝑛
2−∑𝑠∈𝑆′
2𝑐𝑠1 − 𝜆𝑠
𝑠
where 𝑥1, ..., 𝑥𝑛 ∈ h*𝑊 ′ is any basis of h*𝑊 ′ and 𝑦1, ..., 𝑦𝑛 is the dual basis of h𝑊 ′ .
Recall that h𝑊 ′ satisfies the commutation relations [h𝑊 ′ , 𝑥] = 𝑥 for all 𝑥 ∈ h*𝑊 ′ ,
[h𝑊 ′ , 𝑦] = −𝑦 for all 𝑦 ∈ h𝑊 ′ , and [h𝑊 ′ , 𝑤] = 0 for all 𝑤 ∈ 𝑊 . Note that in our setting,
h𝑊 ′ is also fixed by the action of 𝑁𝑊 (𝑊 ′) on 𝐻𝑐(𝑊′, h𝑊 ′) because the parameter
𝑐 : 𝑆 ′ → C is obtained by restriction of the 𝑊 -invariant parameter 𝑐 : 𝑆 → C.
Let 𝒪(𝑁𝑊 (𝑊 ′)n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′)⊗𝐷(h𝑊
′𝑟𝑒𝑔))) denote the category of modules over
the algebra
𝑁𝑊 (𝑊 ′) n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′) ⊗𝐷(h𝑊
′
𝑟𝑒𝑔))
that are finitely generated over C[h𝑊′
𝑟𝑒𝑔] ⊗C[h𝑊 ′ ] and on which h𝑊 ′ acts locally nilpo-
tently, and similarly let 𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ ( 𝐻𝑐(𝑊′, h𝑊 ′)0⊗𝐷(h𝑊
′𝑟𝑒𝑔))) denote the cat-
egory of 𝑁𝑊 (𝑊 ′) n𝑊 ′ ( 𝐻𝑐(𝑊′, h𝑊 ′)0⊗𝐷(h𝑊
′𝑟𝑒𝑔))-modules that are finitely generated
over C[h𝑊′
𝑟𝑒𝑔]⊗C[[h𝑊 ′ ]]. By the proof of [5, Proposition 2.4], h𝑊 ′ acts locally finitely on
29
any 𝐻𝑐(𝑊′, h𝑊 ′)-module that has locally nilpotent action of h𝑊 ′ , and in particular
h𝑊 ′ acts locally finitely on any 𝑀 ∈ 𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′) ⊗ 𝐷(h𝑊
′𝑟𝑒𝑔))). It
follows in particular that such 𝑀 are naturally graded as
𝑀 =⨁𝑎∈C
𝑀𝑎
where 𝑀𝑎 is the generalized 𝑎-eigenspace of h𝑊 ′ in 𝑀 . As the actions of h𝑊 ′ and
𝑁𝑊 (𝑊 ′) n𝐷(h𝑊′
𝑟𝑒𝑔) on 𝑀 commute, it follows that the above decomposition of 𝑀 is
a decomposition as 𝑁𝑊 (𝑊 ′)×𝐷(h𝑊′
𝑟𝑒𝑔)-modules. As h𝑊 ′𝑀𝑎 ⊂𝑀𝑎+1 for all 𝑎 ∈ C and
as 𝑀 is finitely generated over C[h𝑊 ′ ] ⊗ C[h𝑊′
𝑟𝑒𝑔], it follows that each 𝑀𝑎 is finitely
generated over C[h𝑊′
𝑟𝑒𝑔] and that the set of 𝑎 ∈ C such that 𝑀𝑎 = 0 is a subset of a
finite union of sets of the form 𝑧+Z≥0. In particular, the generalized eigenspaces 𝑀𝑎
are 𝑁𝑊 (𝑊 ′)-equivariant 𝒪(h𝑊′
𝑟𝑒𝑔)-coherent 𝐷(h𝑊′
𝑟𝑒𝑔)-modules. An argument entirely
analogous to the proof of Theorem 2.3 in [5] then shows that the functor
𝐸 : 𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ ( 𝐻𝑐(𝑊′, h𝑊 ′)0⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))
→ 𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′) ⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))
sending a module 𝑀 to its subspace of h𝑊 ′-locally finite vectors is quasi-inverse to
the functor h𝑊 ′𝑟𝑒𝑔
: 𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′) ⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))
→ 𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ ( 𝐻𝑐(𝑊′, h𝑊 ′)0⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))
of completion at h𝑊′
𝑟𝑒𝑔.
Let 𝒪( 𝐻𝑐(𝑊, h)𝑊h𝑊′
𝑟𝑒𝑔) denote the category of 𝐻𝑐(𝑊, h)𝑊h𝑊
′𝑟𝑒𝑔
-modules finitely gen-
erated over the ring of functions C[h]∧𝑊h𝑊
′𝑟𝑒𝑔 on the formal neighborhood of 𝑊h𝑊
′𝑟𝑒𝑔 in
h. Completion at 𝑊h𝑊′ defines an exact functor
𝑊h𝑊′
𝑟𝑒𝑔: 𝒪𝑐(𝑊, h) → 𝒪( 𝐻𝑐(𝑊, h)𝑊h𝑊
′𝑟𝑒𝑔
).
30
Note also that the discussion above and the observation that 1h𝑊′
𝑟𝑒𝑔generates the unit
ideal in 𝐻𝑐(𝑊, h)𝑊h𝑊′
𝑟𝑒𝑔shows that the composition
𝜓* ∘ 1h𝑊′
𝑟𝑒𝑔: 𝒪( 𝐻𝑐(𝑊, h)𝑊h𝑊
′𝑟𝑒𝑔
) → 𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ ( 𝐻𝑐(𝑊′, h𝑊 ′)0⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))
is an equivalence of categories. Next, consider the composite functor
𝐸 ∘ 𝜓* ∘ 1h𝑊′
𝑟𝑒𝑔∘ 𝑊h𝑊
′𝑟𝑒𝑔
: 𝒪𝑐(𝑊, h) → 𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′) ⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))
which as we’ve seen is a completion functor followed by equivalences of categories.
Lemma 2.5.0.1. The image of the composite functor 𝐸 ∘ 𝜓* ∘ 1h𝑊′
𝑟𝑒𝑔∘ 𝑊h𝑊
′𝑟𝑒𝑔
lies in
the full subcategory
𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′) ⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))𝑟.𝑠.
of
𝒪(𝑁𝑊 (𝑊 ′) n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′) ⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))
consisting of those modules 𝑀 whose generalized h𝑊 ′ eigenspaces 𝑀𝑎 have regular
singularities as 𝒪(h𝑊′
𝑟𝑒𝑔)-coherent 𝐷(h𝑊′
𝑟𝑒𝑔)-modules.
Proof. Let 𝑁 be a module in 𝒪𝑐(𝑊, h) and let 𝑀 be its image under the composite
functor 𝐸 ∘ 𝜓* ∘ 1h𝑊′
𝑟𝑒𝑔∘ 𝑊h𝑊
′𝑟𝑒𝑔
. As the full subcategory of 𝒪(h𝑊′
𝑟𝑒𝑔)-coherent 𝐷(h𝑊′
𝑟𝑒𝑔)-
modules consisting of those 𝐷-modules with regular singularities is a Serre subcat-
egory, to show that the 𝐷(h𝑊′
𝑟𝑒𝑔)-module 𝑀𝑎 has regular singularities it suffices to
show that every irreducible 𝐷(h𝑊′
𝑟𝑒𝑔)-module appearing as a composition factor in 𝑀
has regular singularities. Note that h*𝑊 ′ acts on 𝑀 by 𝐷(h𝑊′
𝑟𝑒𝑔)-module homomor-
phisms of degree 1 with respect to the h𝑊 ′-grading. It follows that (h*𝑊 ′)𝑘𝑀 is a
𝐷(h𝑊′
𝑟𝑒𝑔)-submodule of 𝑀 for all integers 𝑘 ≥ 0, and, as 𝑀 is finitely generated over
C[h𝑊 ′ ]⊗C[h𝑊′
𝑟𝑒𝑔], that any generalized h𝑊 ′-eigenspace 𝑀𝑎 of 𝑀 embeds in some quo-
tient 𝑀/(h*𝑊 ′)𝑘𝑀 for sufficiently large 𝑘. Therefore, it suffices to show that the
module 𝑀/(h*𝑊 ′)𝑘𝑀 has regular singularities for all integers 𝑘 > 0. But 𝑀/h*𝑊 ′𝑀 is
31
precisely the 𝐷(h𝑊′
𝑟𝑒𝑔)-module obtained by pulling 𝑁 back to h𝑊′
𝑟𝑒𝑔 as a coherent sheaf
as in [73, Proposition 1.2], which has regular singularities by [73, Proposition 1.3].
That 𝑀/(h*𝑊 ′)𝑘𝑀 has regular singularities then follows from the exact sequence
0 → (h*𝑊 ′)𝑘−1𝑀
(h*𝑊 ′)𝑘𝑀→ 𝑀
(h*𝑊 ′)𝑘𝑀→ 𝑀
(h*𝑊 ′)𝑘−1𝑀→ 0
and induction on 𝑘.
By the Riemann-Hilbert correspondence [16], passing to local systems of flat sec-
tions on h𝑊′
𝑟𝑒𝑔 in each generalized eigenspace then defines an equivalence of categories
𝑅𝐻 : 𝒪(𝑁𝑊 (𝑊 ′)n𝑊 ′ (𝐻𝑐(𝑊′, h𝑊 ′)⊗𝐷(h𝑊
′
𝑟𝑒𝑔)))𝑟.𝑠. → (𝒪𝑐(𝑊′, h𝑊 ′)Loc(h𝑊
′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′)
where 𝒪𝑐(𝑊′, h𝑊 ′)Loc(h𝑊
′𝑟𝑒𝑔) is the category of local systems on h𝑊
′𝑟𝑒𝑔 of 𝐻𝑐(𝑊
′, h𝑊 ′)-
modules in 𝒪𝑐(𝑊′, h𝑊 ′) and where (𝒪𝑐(𝑊
′, h𝑊 ′)Loc(h𝑊′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′) is the associated
category of 𝑁𝑊 (𝑊 ′)-equivariant objects.
Definition 2.5.0.2. Let Res𝑊𝑊 ′ be the partial KZ functor
Res𝑊𝑊 ′ : 𝒪𝑐(𝑊, h) → (𝒪𝑐(𝑊′, h𝑊 ′) Loc(h𝑊
′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′)
defined by the composition
Res𝑊𝑊 ′ := 𝑅𝐻 ∘ 𝐸 ∘ 𝜓* ∘ 1h𝑊′
𝑟𝑒𝑔∘ 𝑊h𝑊
′𝑟𝑒𝑔.
Let 𝑏 ∈ h𝑊′
𝑟𝑒𝑔 and let Fiber𝑏 : (𝒪𝑐(𝑊′, h𝑊 ′) Loc(h𝑊
′𝑟𝑒𝑔))
𝑁𝑊 (𝑊 ′) → 𝒪𝑐(𝑊′, h𝑊 ′)
denote the exact functor of taking the fiber of the local system at 𝑏. From [5, Section
3.7], the functor Fiber𝑏 ∘ Res𝑊𝑊 ′ is naturally identified with the parabolic restriction
functor Res𝑏. In particular, the monodromy in h𝑊′
𝑟𝑒𝑔 provides isomorphisms of functors
Res𝑏 ∼= Res𝑏′ for any points 𝑏, 𝑏′ ∈ h𝑊′
𝑟𝑒𝑔. Note, however, that such an isomorphism is
not canonical and depends on a choice of path between 𝑏 and 𝑏′. This monodromy
action on Res𝑏 will be crucial in what follows.
When there is no loss of clarity, we will suppress the choice of 𝑏 ∈ h𝑊′ and write
32
Res𝑊𝑊 ′ for the parabolic restriction functor Res𝑏, and similarly we will write Ind𝑊𝑊 ′ for
the parabolic induction functor Ind𝑏. The underlined Res𝑊𝑊 ′ will always denote the
associated partial 𝐾𝑍 functor with fibers Res𝑊𝑊 ′ .
2.6 Gordon-Martino Transitivity
Let 𝑊 ′′ ⊂ 𝑊 ′ ⊂ 𝑊 be a chain of parabolic subgroups of 𝑊 . In [65, Corollary 2.5],
Shan proved that the parabolic restriction functors are transitive in the sense that
there is an isomorphism of functors
Res𝑊𝑊 ′′ ∼= Res𝑊′
𝑊 ′′ ∘ Res𝑊𝑊 ′ .
Gordon and Martino [40] explained the sense in which this transitivity is compatible
with the local systems appearing in the partial 𝐾𝑍 functors. We recall their result
below.
Consider the functor
Res𝑊′
𝑊 ′′ Id∘ ↓𝑁𝑊 (𝑊 ′)𝑁𝑊 (𝑊 ′,𝑊 ′′) ∘Res𝑊𝑊 ′ : 𝒪𝑐(𝑊, h)
→(𝒪𝑐(𝑊
′′, h𝑊 ′′) Loc((h𝑊 ′)𝑊′′
𝑟𝑒𝑔 ) Loc(h𝑊′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′,𝑊 ′′)
.
Here 𝑁𝑊 (𝑊 ′,𝑊 ′′) is the intersection of the normalizers 𝑁𝑊 (𝑊 ′) and 𝑁𝑊 (𝑊 ′′) and
↓ denotes the restriction of equivariant structure to a subgroup. Similarly to the
notation h𝑊′
𝑟𝑒𝑔, the space (h𝑊 ′)𝑊′′
𝑟𝑒𝑔 is the locally closed locus of points in h𝑊 ′ with
stabilizer 𝑊 ′′ in 𝑊 ′. The goal is to relate this functor to the partial 𝐾𝑍 functor
Res𝑊𝑊 ′′ . However, the latter functor produces local systems on the space h𝑊′′
𝑟𝑒𝑔 , not on
(h𝑊 ′)𝑊′′
𝑟𝑒𝑔 × h𝑊′
𝑟𝑒𝑔 as the functor considered above does. In general, viewed as subvari-
eties of h, these spaces do not coincide, and there is no obvious map between them.
However, (h𝑊 ′)𝑊′′
𝑟𝑒𝑔 ×h𝑊′
𝑟𝑒𝑔, viewed as a complex manifold, does contain a 𝑁𝑊 (𝑊 ′,𝑊 ′′)-
stable deformation retract that includes naturally in h𝑊′′
𝑟𝑒𝑔 , giving rise to a pullback
33
functor
𝜄*𝑊 ′,𝑊 ′′ : Loc(h𝑊′′
𝑟𝑒𝑔 )𝑁𝑊 (𝑊 ′,𝑊 ′′) → Loc((h𝑊 ′)𝑊′′
𝑟𝑒𝑔 × h𝑊′
𝑟𝑒𝑔)𝑁𝑊 (𝑊 ′,𝑊 ′′).
This functor can be constructed as follows.
Choose a 𝑊 -invariant norm || · || on h. Let 𝜖 > 0 be a positive number. Define
the subspaces
h𝑊′
𝜖−𝑟𝑒𝑔 := 𝑥 ∈ h𝑊′: ||𝑤𝑥− 𝑥|| > 𝜖 for all 𝑤 ∈ 𝑊∖𝑊 ′ ⊂ h𝑊
′
𝑟𝑒𝑔
and
(h𝑊 ′)𝑊′′,𝜖
𝑟𝑒𝑔 := 𝑥 ∈ (h𝑊 ′)𝑊′′
𝑟𝑒𝑔 : ||𝑥|| < 𝜖/2 ⊂ (h𝑊 ′)𝑊′′
𝑟𝑒𝑔 .
Note that these spaces are 𝑁𝑊 (𝑊 ′,𝑊 ′′)-stable, that the subspace (h𝑊 ′)𝑊′′,𝜖
𝑟𝑒𝑔 × h𝑊′
𝜖−𝑟𝑒𝑔
is a deformation retract of (h𝑊 ′)𝑊′′
𝑟𝑒𝑔 ×h𝑊′
𝑟𝑒𝑔 via a 𝑁𝑊 (𝑊 ′,𝑊 ′′)-equivariant deformation
retraction, and there is a natural inclusion (h𝑊 ′)𝑊′′,𝜖
𝑟𝑒𝑔 × h𝑊′
𝜖−𝑟𝑒𝑔 ⊂ h𝑊′′
𝑟𝑒𝑔 . Pullback along
the inclusion (h𝑊 ′)𝑊′′,𝜖
𝑟𝑒𝑔 × h𝑊′
𝜖−𝑟𝑒𝑔 ⊂ h𝑊′′
𝑟𝑒𝑔 defines a functor
Loc(h𝑊′′
𝑟𝑒𝑔 )𝑁𝑊 (𝑊 ′,𝑊 ′′) → Loc((h𝑊 ′)𝑊′′,𝜖
𝑟𝑒𝑔 × h𝑊′
𝜖−𝑟𝑒𝑔)𝑁𝑊 (𝑊 ′,𝑊 ′′)
and the deformation retraction defines an equivalence of categories
Loc((h𝑊 ′)𝑊′′,𝜖
𝑟𝑒𝑔 × h𝑊′
𝜖−𝑟𝑒𝑔)𝑁𝑊 (𝑊 ′,𝑊 ′′) ∼= Loc((h𝑊 ′)𝑊
′′
𝑟𝑒𝑔 × h𝑊′
𝑟𝑒𝑔)𝑁𝑊 (𝑊 ′,𝑊 ′′).
We define 𝜄*𝑊 ′,𝑊 ′′ to be the composition of these functors. Note that 𝜄*𝑊 ′,𝑊 ′′ does not
depend on the choice of norm || · || or 𝜖 > 0. We comment that the definition of 𝜄*𝑊 ′,𝑊 ′′
given in [40] was stated in terms of fundamental groups, but the version we give here
is equivalent.
We can now state the transitivity result of Gordon-Martino:
Theorem 2.6.0.1. (Gordan-Martino, [40, Theorem 3.11]) There is a natural isomor-
34
phism
𝜄*𝑊 ′,𝑊 ′′∘ ↓𝑁𝑊 (𝑊 ′′)𝑁𝑊 (𝑊 ′,𝑊 ′′) ∘Res𝑊𝑊 ′′ ∼= (Res𝑊
′
𝑊 ′′ Id)∘ ↓𝑁𝑊 (𝑊 ′)𝑁𝑊 (𝑊 ′,𝑊 ′′) ∘Res𝑊𝑊 ′
of functors
𝒪𝑐(𝑊, h) →(𝒪𝑐(𝑊
′′, h𝑊 ′′) Loc((h𝑊 ′)𝑊′′
𝑟𝑒𝑔 ) Loc(h𝑊′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′,𝑊 ′′)
.
2.7 Mackey Formula for Rational Cherednik Alge-
bras for Coxeter Groups
Shan and Vasserot established in [66, Lemma 2.5] that the natural analogue of the
usual Mackey formula for the composition of induction and restriction functors for
representations of finite groups holds for the Bezrukavnikov-Etingof parabolic induc-
tion and restriction functors at the level of Grothendieck groups. It will be convenient
for us to know that, at least in the case of rational Cherednik algebras attached to
finite Coxeter groups, the Mackey formula holds at the level of the parabolic induction
and restriction functors themselves. This is established below, by lifting the Mackey
formula for the associated Hecke algebras via the 𝐾𝑍 functor.
Let (𝑊,𝑆) be a finite Coxeter system with simple reflections 𝑆, real reflection
representation hR, and complexified reflection representation h. Let 𝑐 : 𝑆 → C be a
class function on the simple reflections, and let 𝑞 : 𝑆 → C× be the associated class
function 𝑞(𝑠) = 𝑒−2𝜋𝑖𝑐(𝑠). Recall that the Hecke algebra H𝑞(𝑊 ) attached to 𝑊 and
𝑞 is the associative C-algebra generated by symbols 𝑇𝑠 : 𝑠 ∈ 𝑆 subject to braid
relations
𝑇𝑠1𝑇𝑠2𝑇𝑠1 · · · = 𝑇𝑠2𝑇𝑠1𝑇𝑠2 · · ·
for 𝑠1 = 𝑠2 ∈ 𝑆, where there are 𝑚𝑖𝑗 terms on each side where 𝑚𝑖𝑗 is the order of the
product 𝑠1𝑠2, and the quadratic relations
(𝑇𝑠 − 1)(𝑇𝑠 + 𝑞𝑠) = 0
35
where we write 𝑞𝑠 for 𝑞(𝑠). Recall that the length function 𝑙 : 𝑊 → Z≥0 assigns
to 𝑤 ∈ 𝑊 the minimal length 𝑙(𝑤) of an expression 𝑠𝑖1 · · · 𝑠𝑖𝑙(𝑤)= 𝑤 of 𝑤 as a
product of simple reflections, and that such a factorization of 𝑤 is called a reduced
expression. The product 𝑇𝑠𝑖1 · · ·𝑇𝑠𝑖𝑙(𝑤)in H𝑞(𝑊 ) does not depend on the choice of
reduced expression 𝑠𝑖1 · · · 𝑠𝑖𝑙(𝑤)for 𝑊 , and the resulting elements 𝑇𝑤 form a C-basis
for H𝑞(𝑊 ). For details and proofs on such basic structural results on Hecke algebras
mentioned in this section, see [37] (note that their convention is to use quadratic
relations (𝑇 + 1)(𝑇 − 𝑞) = 0 rather than (𝑇 − 1)(𝑇 + 𝑞) = 0).
For a subset 𝐽 ⊂ 𝑆 of the simple reflections in 𝑊 , let 𝑊𝐽 ⊂ 𝑊 be the parabolic
subgroup generated by 𝐽 ⊂ 𝑆. Then (𝑊𝐽 , 𝐽) is a Coxeter subsystem of (𝑊,𝑆), and
the (complexified) reflection representation of (𝑊𝐽 , 𝐽) is identified with the action of
𝑊𝐽 on the unique 𝑊𝐽 -stable complement h𝑊𝐽of h𝑊𝐽 in h. The parameter 𝑞 : 𝑆 → C×
restricts to give a 𝑊𝐽 -invariant function 𝐽 → C×, which by abuse of notation we also
denote by 𝑞. We therefore may form the Hecke algebra H𝑞(𝑊𝐽). Reduced expressions
of 𝑤 ∈ 𝑊𝐽 as an element of 𝑊𝐽 are precisely the reduced expressions of 𝑤 as an
element of 𝑊 . In particular the length function 𝑙𝑊𝐽for (𝑊𝐽 , 𝐽) coincides with the
restriction to 𝑊𝐽 of the length function for (𝑊,𝐽), and in this way the assignment
𝑇𝑤 ↦→ 𝑇𝑤 for 𝑤 ∈ 𝑊 extends uniquely to a unital embedding of the algebra H𝑞(𝑊𝐽)
as a subalgebra of H𝑞(𝑊 ). Via this embedding we can define the parabolic restriction
functor𝐻Res𝑊𝑊𝐽
: H𝑞(𝑊 )-mod → H𝑞(𝑊𝐽)-mod
and the parabolic induction functor
𝐻Ind𝑊𝑊𝐽:= H𝑞(𝑊 ) ⊗H𝑞(𝑊𝐽 ) ∙ : H𝑞(𝑊𝐽)-mod → H𝑞(𝑊 )-mod.
Let 𝐽,𝐾 ⊂ 𝑆 be two subsets of simple reflections, with associated parabolic sub-
groups 𝑊𝐽 and 𝑊𝐾 . Each 𝑊𝐾-𝑊𝐽 double-coset in 𝑊 contains a unique element
of minimal length. Let 𝑋𝐾𝐽 ⊂ 𝑊 denote this subset of distinguished double-coset
representatives. For 𝑑 ∈ 𝑋𝐾𝐽 , conjugation 𝑤 ↦→ 𝑑𝑤𝑑−1 defines a length-preserving
isomorphism 𝑊𝐽∩𝐾𝑑 → 𝑊𝑑𝐽∩𝐾 , where 𝐾𝑑 := 𝑑−1𝐾𝑑 and 𝑑𝐽 := 𝑑𝐽𝑑−1. In par-
36
ticular, the assignment 𝑇𝑤 ↦→ 𝑇𝑑𝑤𝑑−1 extends uniquely to an algebra isomorphism
H𝑞(𝑊𝐽∩𝐾𝑑) → H𝑞(𝑊𝑑𝐽∩𝐾). Note that this isomorphism is realized inside H𝑞(𝑊 ) by
conjugation by 𝑇𝑑. Let
𝐻tw𝑑 : H𝑞(𝑊𝐽∩𝐾𝑑)-mod → H𝑞(𝑊𝑑𝐽∩𝐾)-mod
denote the equivalence of categories obtained by transfer of structure via this isomor-
phism.
We can now state the Mackey formula for Hecke algebras. Note that this formula
holds for any numerical parameter 𝑞 : 𝑆 → C×.
Proposition 2.7.0.1. (Mackey Formula for Hecke Algebras, [37, Proposition 9.1.8])
Let 𝐽,𝐾 ⊂ 𝑆 and let 𝑋𝐾𝐽 ⊂ 𝑊 be the set of distinguished (minimal length) 𝑊𝐾-𝑊𝐽
double-coset representatives in 𝑊 . There is an isomorphism
𝐻Res𝑊𝑊𝐾∘ 𝐻Ind𝑊𝑊𝐽
∼=⨁𝑑∈𝑋𝐾𝐽
𝐻Ind𝑊𝑊𝑑𝐽∩𝐾∘ 𝐻tw𝑑 ∘ 𝐻Res𝑊𝐽
𝐽∩𝐾𝑑
of functors H𝑞(𝑊𝐽)-mod → H𝑞(𝑊𝐾)-mod.
By restriction of the parameter 𝑐 to subsets 𝐽 ⊂ 𝑆, one can form the associated ra-
tional Cherednik algebra 𝐻𝑐(𝑊𝐽 , h𝑊𝐽). Let 𝐽,𝐾 ⊂ 𝑆, and let 𝑑 ∈ 𝑋𝐾𝐽 . Note that the
action of 𝑑 on h induces an isomorphism h𝑊𝐽∩𝐾𝑑
→ h𝑊𝑑𝐽∩𝐾, 𝑦 ↦→ 𝑑𝑦, and hence also an
isomorphism h*𝑊𝐽∩𝐾𝑑
→ h*𝑊𝑑𝐽∩𝐾, 𝑓 ↦→ 𝑑(𝑓) = 𝑓(𝑑−1∙). As in the comments preceding
[66, Lemma 2.5], these isomorphisms along with the isomorphism 𝑊𝐽∩𝐾𝑑 → 𝑊𝑑𝐽∩𝐾
discussed above induce an isomorphism 𝐻𝑐(𝑊𝐽∩𝐾𝑑 , h𝑊𝐽∩𝐾𝑑
) ∼= 𝐻𝑐(𝑊𝑑𝐽∩𝐾 , h𝑊𝑑𝐽∩𝐾) re-
specting triangular decompositions. Transfer of structure by this isomorphism there-
fore induces an equivalence of categories
tw𝑑 : 𝒪𝑐(𝑊𝐽∩𝐾𝑑 , h𝑊𝐽∩𝐾𝑑
) ∼= 𝒪𝑐(𝑊𝑑𝐽∩𝐾 , h𝑊𝑑𝐽∩𝐾).
Theorem 2.7.0.2. (Mackey Formula for Rational Cherednik Algebras for Coxeter
Groups) Let 𝐽,𝐾 ⊂ 𝑆 and let 𝑋𝐾𝐽 ⊂ 𝑊 be the set of distinguished 𝑊𝐾-𝑊𝐽 double-
37
coset representatives in 𝑊 . There is an isomorphism
Res𝑊𝑊𝐾∘ Ind𝑊𝑊𝐽
∼=⨁𝑑∈𝑋𝐾𝐽
Ind𝑊𝑊𝑑𝐽∩𝐾∘ tw𝑑 ∘ Res𝑊𝐽
𝐽∩𝐾𝑑
of functors 𝒪𝑐(𝑊𝐽 , h𝑊𝐽) → 𝒪𝑐(𝑊𝐾 , h𝑊𝐾
).
Proof. For a subset 𝐴 ⊂ 𝑆, let 𝐾𝑍(𝑊𝐴, h𝑊𝐴) denote the 𝐾𝑍 functor 𝒪𝑐(𝑊𝐴, h𝑊𝐴
) →
H𝑞(𝑊𝐴)-mod. It follows from [65, Lemma 2.4] that to show the existence of the desired
isomorphism of functors, we need only check that the functors𝐾𝑍(𝑊𝐾 , h𝑊𝐾 )∘Res𝑊𝑊𝐾
∘
Ind𝑊𝑊𝐽and 𝐾𝑍(𝑊𝐾 , h𝑊𝐾
) ∘⨁
𝑑∈𝑋𝐾𝐽Ind𝑊𝑊𝑑𝐽∩𝐾
∘ tw𝑑 ∘ Res𝑊𝐽
𝐽∩𝐾𝑑 are isomorphic. Shan
also proved ([65, Theorem 2.1]) that𝐾𝑍 commutes with parabolic restriction functors
in the sense that there is an isomorphism of functors
𝐾𝑍(𝑊𝐴, h𝑊𝐴) ∘ Res𝑊𝐵
𝑊𝐴
∼= 𝐻Res𝑊𝐵𝑊𝐴
∘𝐾𝑍(𝑊𝐵, h𝑊𝐵)
for any 𝐴 ⊂ 𝐵 ⊂ 𝑆. It is clear that 𝐾𝑍 commutes with tw𝑑 in the sense that there is
an isomorphism of functors 𝐾𝑍(𝑊𝑑𝐽∩𝐾 , h𝑊 𝑑𝐽∩𝐾)∘tw𝑑∼= 𝐻tw𝑑∘𝐾𝑍(𝑊𝐽∩𝐾𝑑 , h𝑊
𝐽∩𝐾𝑑).
Passing the 𝐾𝑍 functor to the right using these isomorphisms, the desired statement
then follows from the Mackey formula for Hecke algebras.
38
Chapter 3
The Refined Filtration By Supports
3.1 Introduction
A key technical tool for studying 𝐻𝑐(𝑊, h) and the category 𝒪𝑐(𝑊, h) is the Knizhnik-
Zamolodchikov (KZ) functor also introduced in [38]. The KZ functor
𝐾𝑍 : 𝒪𝑐(𝑊, h) → H𝑞(𝑊 )-mod𝑓.𝑑.
is an exact functor, defined via monodromy, from 𝒪𝑐(𝑊, h) to the category
H𝑞(𝑊 )-mod𝑓.𝑑.
of finite-dimensional modules over the Hecke algebra H𝑞(𝑊 ) associated to the reflec-
tion group𝑊 . The parameter 𝑞 of the Hecke algebra H𝑞(𝑊 ) depends on the parameter
𝑐 of the rational Cherednik algebra 𝐻𝑐(𝑊, h) in an exponential manner. 𝐾𝑍 induces
an equivalence of categories 𝒪𝑐(𝑊, h)/𝒪𝑐(𝑊, h)𝑡𝑜𝑟 ∼= H𝑞(𝑊 )-mod𝑓.𝑑 [38, 53], where
𝒪𝑐(𝑊, h)𝑡𝑜𝑟 denotes the Serre subcategory of modules in 𝒪𝑐(𝑊, h) supported on the
union of the reflection hyperplanes ∪𝑠∈𝑆 ker(𝑠). In this way, 𝐾𝑍 establishes a bi-
jection between the irreducible representations in 𝒪𝑐(𝑊, h) of full support in h and
the finite-dimensional irreducible representations of H𝑞(𝑊 ). Following previous work
of Etingof-Rains [32], Marin-Pfeiffer [57], Losev [53], Chavli [9], and others towards
39
proving the Broué-Malle-Rouquier conjecture [7], Etingof [27] recently showed that
the Hecke algebra H𝑞(𝑊 ) is always finite-dimensional with dimension #𝑊 , even in
the case of complex reflection groups.
In this chapter, in the case that 𝑊 is a Coxeter group with complexified reflection
representation h, we extend this correspondence between irreducible representations
𝐿 in 𝒪𝑐(𝑊, h) and irreducible representations of finite-type Hecke algebras to include
all cases in which the support of 𝐿 is not equal to 0 ⊂ h, i.e. all cases in which 𝐿 is
not finite-dimensional over C. Our approach is inspired by the Harish-Chandra series
appearing in the representation theory of finite groups of Lie type. In place of the
parabolic induction and restriction functors defined for finite groups of Lie type, in
the setting of rational Cherednik algebras one has analogous parabolic induction and
restriction functors introduced by Bezrukavnikov and Etingof [5]. In particular, sup-
pose (𝑊 ′, 𝑆 ′) ⊂ (𝑊,𝑆) is a parabolic Coxeter subsystem of (𝑊,𝑆) with complexified
reflection representation h𝑊 ′ . Restricting the parameter 𝑐 to 𝑆 ′ we may form the ra-
tional Cherednik algebra 𝐻𝑐(𝑊′, h𝑊 ′) and the associated category of representations
𝒪𝑐(𝑊′, h𝑊 ′). The algebra 𝐻𝑐(𝑊
′, h𝑊 ′) does not naturally embed as a subalgebra of
the algebra 𝐻𝑐(𝑊, h) as is the case for Hecke algebras, so there is no naive definition of
parabolic induction and restriction functors as there is for Hecke algebras. Rather, the
Bezrukavinikov-Etingof parabolic induction functor Ind𝑊𝑊 ′ : 𝒪𝑐(𝑊′, h𝑊 ′) → 𝒪𝑐(𝑊, h)
and restriction functor Res𝑊𝑊 ′ : 𝒪𝑐(𝑊, h) → 𝒪𝑐(𝑊′, h𝑊 ′) are more technical and are
defined using the geometric interpretation of rational Cherednik algebras and a cer-
tain isomorphism involving completions of 𝐻𝑐(𝑊, h) and 𝐻𝑐(𝑊′, h𝑊 ′). The functors
Res𝑊𝑊 ′ and Ind𝑊𝑊 ′ are exact [5] and biadjoint [54, 65].
As for representations of finite groups of Lie type, one defines the cuspidal rep-
resentations of a rational Cherednik algebra 𝐻𝑐(𝑊, h) to be those irreducible repre-
sentations 𝐿 in 𝒪𝑐(𝑊, h) such that Res𝑊𝑊 ′𝐿 = 0 for all proper parabolic subgroups
𝑊 ′ ⊂ 𝑊 . For rational Cherednik algebras, the cuspidal representations are precisely
the finite-dimensional irreducible representations. The class of cuspidal representa-
tions is the smallest class of representations in the categories 𝒪𝑐(𝑊, h) such that any
irreducible representation in a category 𝒪𝑐(𝑊, h) appears as a subobject (or, as a
40
quotient) of a representation induced from a representation in that class. From this
perspective, the study of irreducible representations in 𝒪𝑐(𝑊, h) is largely reduced
to the study of cuspidal representations 𝐿 in categories 𝒪𝑐(𝑊′, h𝑊 ′) for 𝑊 ′ ⊂ 𝑊 a
parabolic subgroup and of the structure and decomposition of the induced represen-
tations Ind𝑊𝑊 ′𝐿.
The most basic finite-dimensional irreducible representation of a rational Chered-
nik algebra is the trivial representation C of the trivial rational Cherednik algebra
𝐻𝑐(1, 0) = C. The induced representation Ind𝑊1 C, denoted 𝑃𝐾𝑍 , is a remark-
able projective object in 𝒪𝑐(𝑊, h). In particular, there is an algebra isomorphism
End𝐻𝑐(𝑊,h)(𝑃𝐾𝑍)𝑜𝑝𝑝 ∼= H𝑞(𝑊 ) with respect to which 𝑃𝐾𝑍 represents the 𝐾𝑍 func-
tor. The equivalence of categories 𝒪𝑐(𝑊, h)/𝒪𝑐(𝑊, h)𝑡𝑜𝑟 ∼= H𝑞(𝑊 )-mod𝑓.𝑑. induced by
𝐾𝑍 therefore establishes a correspondence between the irreducible representations in
𝒪𝑐(𝑊, h) of full support in h and the irreducible representations of End𝐻𝑐(𝑊,h)(Ind𝑊1 C).
In this chapter, we study the endomorphism algebras End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿) of in-
duced cuspidal representations 𝐿 in the general case that 𝐿 is not necessarily the triv-
ial representation C in 𝒪𝑐(1, 0) and provide results analogous to and generalizing
those summarized above for the case 𝐿 = C. In the parallel setting of representations
of finite groups of Lie type, endomorphism algebras of induced cuspidal representa-
tions have been studied in great detail and are closely related to Hecke algebras of
finite type. For example, in the most basic case one considers the parabolic induction
Ind𝐺𝐵C of the trivial representation C of the trivial group 1 to the finite general linear
group 𝐺 := 𝐺𝐿𝑛(F𝑞), where 𝑞 is a prime power and 𝐵 is the standard Borel subgroup
of upper triangular matrices. In that case, the endomorphism algebra is precisely
the Hecke algebra H𝑞(𝑆𝑛) where 𝑞 is the order of the finite field, exactly analogous
to the case of 𝑃𝐾𝑍 = Ind𝑊1 C for rational Cherednik algebras. In the general case,
Howlett and Lehrer [46, Theorem 4.14] showed that, in characteristic 0, the endo-
morphism algebra of a parabolically induced cuspidal representation of a finite group
of Lie type can be described as a semidirect product of a finite type Hecke algebra
by a finite group acting by a diagram automorphism, twisted by a certain 2-cocycle
(see, for instance, [8, Theorem 10.8.5]). We obtain an exactly analogous result for
41
the endomorphism algebras of induced representations Ind𝑊𝑊 ′𝐿 of finite-dimensional
irreducible representations 𝐿 of 𝐻𝑐(𝑊′, h𝑊 ′). Howlett and Lehrer conjectured [46,
Conjecture 6.3] that the 2-cocycle appearing in the description of the endomorphism
algebra was trivial, as was later proved (see Lusztig [56, Theorem 8.6] and Geck [35]).
Appealing to the classification of irreducible finite Coxeter groups and to previous
results about finite-dimensional irreducible representations of rational Cherednik al-
gebras, we find that the 2-cocycles appearing in our setting are trivial as well. The
connections between the categories 𝒪𝑐(𝑊, h) and the modular representation theory
of finite groups of Lie type is more than an analogy; Norton [61] explains various facts
and conjectures relating these two subjects, as well as ways in which these connections
break down.
The methods used in the setting of finite groups of Lie type ultimately rely on
the comparatively simple and explicit definition of parabolic induction, which allows
for a basis of intertwining operators to be written down in closed form. Instead,
in our setting of the more complicated Bezrukavnikov-Etingof parabolic induction
functors for rational Cherednik algebras, we consider for each finite-dimensional ir-
reducible representation 𝐿 in 𝒪𝑐(𝑊′, h𝑊 ′) the functor 𝐾𝑍𝐿 represented by Ind𝑊𝑊 ′𝐿,
analogous to the 𝐾𝑍 functor. Recall that 𝐾𝑍 induces an equivalence of categories
𝒪𝑐(𝑊, h)/𝒪𝑐(𝑊, h)𝑡𝑜𝑟 ∼= H𝑞(𝑊 )-mod𝑓.𝑑.; the functor 𝐾𝑍𝐿 induces an analogous equiv-
alence. Specifically, let 𝒪𝑐(𝑊, h)(𝑊 ′,𝐿) be the Serre subcategory of 𝒪𝑐(𝑊, h) generated
by those irreducible representations 𝑀 supported on 𝑊h𝑊′ as coherent sheaves on h
and with 𝐿 appearing as a composition factor of Res𝑊𝑊 ′𝑀 , and let 𝒪𝑐(𝑊, h)𝑡𝑜𝑟(𝑊 ′,𝐿) be
the kernel of Res𝑊𝑊 ′ in 𝒪𝑐(𝑊, h)(𝑊 ′,𝐿). Then 𝐾𝑍𝐿 induces an equivalence of categories
𝒪𝑐(𝑊, h)(𝑊 ′,𝐿)/𝒪𝑐(𝑊, h)𝑡𝑜𝑟(𝑊 ′,𝐿)∼= End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝-mod𝑓.𝑑.
As with the 𝐾𝑍 functor, we provide a geometric interpretation of the functor 𝐾𝑍𝐿
in terms of the monodromy of an equivariant local system on h𝑊′
𝑟𝑒𝑔, the open locus
of points inside the fixed space h𝑊′ with stabilizer precisely equal to 𝑊 ′. Using a
transitivity result for local systems of parabolic restriction functors due to Gordon and
42
Martino [40], we reduce the problem of computing the eigenvalues of the monodromy
around the missing hyperplanes, i.e. the parameters of the underlying “Hecke algebra”
End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝, to the case in which 𝑊 ′ is a corank-1 parabolic subgroup of
𝑊 . In that case, we use a different application of the Gordon-Martino transitivity
result and the fact that the usual 𝐾𝑍 functor is fully faithful on projective objects
to further reduce the problem of computing eigenvalues of monodromy to concrete,
although often involved, computations inside the Hecke algebra H𝑞(𝑊 ) itself.
The main technical result of this chapter is the following:
Theorem. Let 𝑊 be a finite Coxeter group with simple reflections 𝑆, let 𝑐 : 𝑆 → C
be a class function, let 𝑊 ′ be a standard parabolic subgroup generated by the simple
reflections 𝑆 ′, and let 𝐿 be an irreducible finite-dimensional representation of the
rational Cherednik algebra 𝐻𝑐(𝑊′, h𝑊 ′). Let 𝑁𝑊 ′ denote the canonical complement
to 𝑊 ′ in its normalizer 𝑁𝑊 (𝑊 ′), let 𝑆𝑊 ′ ⊂ 𝑁𝑊 ′ denote the set of reflections in
𝑁𝑊 ′ with respect to its representation in the fixed space h𝑊′, and let 𝑁 𝑟𝑒𝑓
𝑊 ′ denote the
reflection subgroup of 𝑁𝑊 ′ generated by 𝑆𝑊 ′. Let 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ be a complement for 𝑁 𝑟𝑒𝑓
𝑊 ′ in
𝑁𝑊 ′, given as the stabilizer of a choice of fundamental Weyl chamber for the action
of 𝑁 𝑟𝑒𝑓𝑊 ′ on h𝑊
′. Then there is a class function 𝑞𝑊 ′,𝐿 : 𝑆𝑊 ′ → C× and an isomorphism
of algebras
End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝 ∼= 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ n H𝑞𝑊 ′,𝐿
(𝑁 𝑟𝑒𝑓𝑊 ′ )
where the semidirect product is defined by the action of 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ on H𝑞𝑊 ′,𝐿
(𝑁 𝑟𝑒𝑓𝑊 ′ ) by
diagram automorphisms arising from the conjugation action of 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ on 𝑁 𝑟𝑒𝑓
𝑊 ′ . Fur-
thermore, the parameter 𝑞𝑊 ′,𝐿 is explicitly computable.
The content of the theorem above appears later in Theorem 3.3.2.14, for the
isomorphism End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝 ∼= 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ n H𝑞𝑊 ′,𝐿
(𝑁 𝑟𝑒𝑓𝑊 ′ ), and Theorem 3.3.2.11,
for the computation of 𝑞𝑊 ′,𝐿.
As a corollary of the explicit descriptions of the algebras
End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝
43
we are able to count the number of irreducible representations in 𝒪𝑐(𝑊, h) with any
given support variety in h. In particular, as an application we calculate the number of
finite-dimensional irreducible representations of 𝐻𝑐(𝑊, h) for all exceptional Coxeter
groups 𝑊 and all parameters 𝑐.
This chapter is organized as follows. In Section 3.2, we introduce the filtration
on 𝒪𝑐(𝑊, h) that we study in this chapter. We construct the functor 𝐾𝑍𝐿 using the
monodromy of the Bezrukavnikov-Etingof parabolic restriction functors, providing
a description of the algebra End𝐻𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝 as a quotient of a twisted group
algebra of a fundamental group. We use the Gordon-Martino transitivity result to
show that the generator of monodromy around a deleted hyperplane satisfies a poly-
nomial relation of degree equal to the order of the stabilizer of that hyperplane in the
inertia group of 𝐿 (see Definition 3.2.1.5). In Section 3.3, we specialize to the case in
which 𝑊 is a Coxeter group. In this case, we provide a presentation of the endomor-
phism algebra as a generalized Hecke algebra directly analogous to the presentation
by Howlett and Lehrer [46, Theorem 4.14]. Using the 𝐾𝑍 functor, we provide a
method for computing the parameters of these generalized Hecke algebras. We apply
our methods systematically, using the classification of finite Coxeter groups and pre-
viously known results about finite-dimensional representations, to count the number
of irreducible representations in 𝒪𝑐(𝑊, h) of given support for all exceptional Coxeter
groups 𝑊 and parameters 𝑐. We describe the generalized Hecke algebras appearing
in types 𝐸 and 𝐻 explicitly.
The results we obtain for finite Coxeter groups confirm, unify, and extend many
previously known results in both classical and exceptional types. In type A, we
recover Wilcox’s description [73, Theorem 1.2] of the subquotients of the filtration by
supports of the categories 𝒪𝑐(𝑆𝑛, h). In type 𝐵, we show that the subquotients of our
refined filtration are equivalent to categories of finite-dimensional modules over tensor
products of Iwahori-Hecke algebras of type 𝐵, and we obtain similar descriptions in
type 𝐷. These facts follows from results of Shan-Vasserot [66], although their results
do not generalize to the exceptional types.
In the exceptional types, however, our methods provide many new results. Among
44
the exceptional Coxeter types, complete knowledge of character formulas for all irre-
ducible representations in 𝒪𝑐(𝑊, h) for all parameters 𝑐 is only known for the dihedral
types, treated by Chmutova [13], and type 𝐻3, treated by Balagovic-Puranik [2]. For
parameters 𝑐 = 1/𝑑, where 𝑑 > 2 is an integer dividing a fundamental degree of the
Coxeter group 𝑊 , Norton [61] computed the decomposition matrices for 𝒪𝑐(𝑊, h),
thereby classifying the finite-dimensional irreducible representations and determin-
ing character formulas and supports for all irreducible representations, when 𝑊 is a
Coxeter group of type 𝐸6, 𝐸7, 𝐻4, or 𝐹4. After [52] was finished, I learned that Nor-
ton, in a sequel to [61] then to be announced, had produced complete decomposition
matrices in type 𝐸8 when the denominator 𝑑 of 𝑐 is not 2, 3, 4 or 6, decomposition
matrices for several blocks of 𝒪𝑐(𝐸8, h) when the denominator 𝑑 equals 4 or 6, as well
as decomposition matrices in some unequal parameter cases in type 𝐹4. However,
to our knowledge, the classification of the finite-dimensional representations of ratio-
nal Cherednik algebras remained unknown in types 𝐸6, 𝐸7, 𝐻4 and 𝐹4 at half-integer
parameters, in type 𝐹4 in most unequal parameter cases, and in type 𝐸8 when the
denominator 𝑑 of 𝑐 is 2,3,4, or 6.
Recently, Griffeth-Gusenbauer-Juteau-Lanini [45] produced a necessary condition
for an irreducible representation 𝐿𝑐(𝜆) in 𝒪𝑐(𝑊, h) to be finite-dimensional that ap-
plies to all complex reflection groups 𝑊 and parameters 𝑐. Our results show that this
condition is also sufficient in many of the previously unknown cases mentioned above,
providing complete classifications of the irreducible finite-dimensional representations
of 𝐻𝑐(𝑊, h) in these cases. In combination with previous results of Balagovic-Puranik
[2], Norton [61], in this way we complete the classification of finite-dimensional irre-
ducible representations of 𝐻𝑐(𝑊, h) in the case that 𝑊 is an exceptional Coxeter
group and 𝑐 = 1/𝑑, where 𝑑 is a positive integer possibly equal to 2, except in the
case in which both 𝑊 = 𝐸8 and also 𝑑 = 3 simultaneously. In the case 𝑊 = 𝐸8
and 𝑐 = 1/3, we show that there are 8 non-isomorphic finite-dimensional irreducible
representations of 𝐻1/3(𝐸8, h), and results of [45] rule out all but 9 of the 112 irre-
ducible representations of the Coxeter group 𝐸8 as the potential lowest weights of
these representations. In particular, determining which one of these 9 irreducible rep-
45
resentations is infinite-dimensional will complete the classification in type 𝐸 entirely.
This analysis in types 𝐸,𝐻, and 𝐹 is carried out Sections 3.3.7, 3.3.8, and 3.3.9,
respectively.
In Sections 3.3.7 and 3.3.8, we describe the generalized Hecke algebras arising
from exceptional Coxeter groups 𝑊 of types 𝐸6, 𝐸7, 𝐸8, 𝐻3 and 𝐻4 at all parameters
𝑐 of the form 𝑐 = 1/𝑑, where 𝑑 > 1 is an integer dividing a fundamental degree of 𝑊 .
In Section 3.3.9, we count the number of irreducible representations in 𝒪𝑐1,𝑐2(𝐹4, h)
of given support in h for all parameter values 𝑐1, 𝑐2, including the unequal parameter
case. Along with previous results of Chmutova [13] for dihedral Coxeter groups, this
completes the counting of irreducible representations in 𝒪𝑐(𝑊, h) of given support for
all exceptional Coxeter groups 𝑊 and all, possibly unequal, parameters 𝑐.
These results for parameters 𝑐 = 1/𝑑 can be extended to parameters 𝑐 = 𝑟/𝑑
with 𝑟 > 0 a positive integer relatively prime to 𝑑; the case 𝑟 < 0 is then obtained
by tensoring with the sign character. The reduction from the 𝑐 = 𝑟/𝑑 case to the
𝑐 = 1/𝑑 case can be achieved by results of Rouquier [63], Opdam [62], and Gordon-
Griffeth [39] when 𝑑 ≥ 3. Specifically, a result of Rouquier, as it appears in [39,
Theorem 2.7] after appropriate modifications, implies that for any integer 𝑑 ≥ 3
and integer 𝑟 > 1 relatively prime to 𝑑 there is a bijection 𝜙 on the set Irr(𝑊 )
of the irreducible complex representations of the Coxeter group 𝑊 such that for
all 𝜆 ∈ Irr(𝑊 ) the irreducible representations 𝐿1/𝑑(𝜆) and 𝐿𝑟/𝑑(𝜙(𝜆)) have the same
support in h. In [62] these bijections are calculated, and in [39, Section 2.16] a method
for computing the bijections 𝜙 is given which applies in greater generality. This allows
the reduction of the classification of irreducible representations of given support in
𝒪𝑟/𝑑(𝑊, h) to the classification of irreducible representations of the same given support
in 𝒪1/𝑑(𝑊, h). In particular, as the finite-dimensional representations are those with
support 0 ⊂ h, this provides the same reduction for classifying finite-dimensional
irreducible representations. In type 𝐻3, this analysis has been carried out in this
way by Balagovic-Puranik [2, Section 5.4], and this approach generalizes to the other
Coxeter types. When 𝑑 = 2, our results show that the necessary condition for finite-
dimensionality appearing in [45] is in fact also sufficient for the exceptional types 𝐸,𝐻,
46
and 𝐹 , completing the classification of finite-dimensional irreducible representations
for all half-integer parameters in these cases without the use of such bijections on the
labels of the irreducibles.
3.2 Endomorphism Algebras Via Monodromy
Throughout this section, let 𝑊 be a complex reflection group with reflection rep-
resentation h, let 𝑆 ⊂ 𝑊 be the set of reflections in 𝑊 , and let 𝑐 : 𝑆 → C be a
𝑊 -invariant function. Let 𝐻𝑐(𝑊, h) be the associated rational Cherednik algebra.
In light of the isomorphism 𝐻𝑐(𝑊, h) ∼= 𝐻𝑐(𝑊, h𝑊 ) ⊗𝐷(h𝑊 ), we will always assume
that the fixed space h𝑊 is the trivial subspace. In order to understand and count
irreducible representations in 𝒪𝑐(𝑊, h) with a given support in h, it will be convenient
to consider certain subcategories and subquotient categories of 𝒪𝑐(𝑊, h).
3.2.1 Harish-Chandra Series for Rational Cherednik Algebras
In the following definition, let 𝑊 ′ ⊂ 𝑊 be a parabolic subgroup and let 𝐿 be a
finite-dimensional irreducible representation in 𝒪𝑐(𝑊′, h𝑊 ′).
Definition 3.2.1.1. Let 𝑊 ′ ⊂ 𝑊 be a parabolic subgroup and let 𝐿 be a finite-
dimensional irreducible representation in 𝒪𝑐(𝑊′, h𝑊 ′). Let 𝒪𝑐,𝑊 ′(𝑊, h) be the full sub-
category of 𝒪𝑐(𝑊, h) consisting of modules supported on 𝑊h𝑊′, and let 𝒪𝑐,𝑊 ′,𝐿(𝑊, h)
be the full subcategory of 𝒪𝑐,𝑊 ′(𝑊, h) consisting of modules 𝑀 such that Res𝑊𝑊 ′𝑀 ∈
𝒪𝑐(𝑊′, h𝑊 ′) is semisimple with all irreducible constituents in orbit of 𝐿 for the action
of 𝑁𝑊 (𝑊 ′) on Irr(𝐻𝑐(𝑊′, h𝑊 ′)). Let 𝒪𝑐,𝑊 ′(𝑊, h) and 𝒪𝑐,𝑊 ′,𝐿(𝑊, h) be the respective
quotient categories by the kernel of the exact functor Res𝑊𝑊 ′.
When the ambient reflection representation (𝑊, h) is clear, we will write 𝒪𝑐,𝑊 ′
for 𝒪𝑐,𝑊 ′(𝑊, h), 𝒪𝑐,𝑊 ′,𝐿 for 𝒪𝑐,𝑊 ′,𝐿(𝑊, h), and similarly for their respective quotients
𝒪𝑐,𝑊 ′ and 𝒪𝑐,𝑊 ′,𝐿.
Proposition 3.2.1.2. The following hold:
47
(1) We have
Ext1𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿′, 𝐿′′) = 0
for all 𝐿′, 𝐿′′ in the 𝑁𝑊 (𝑊 ′)-orbit of 𝐿.
(2) The categories 𝒪𝑐,𝑊 ′ and 𝒪𝑐,𝑊 ′,𝐿 are Serre subcategories of 𝒪𝑐(𝑊, h).
(3) Every simple object 𝑆 ∈ 𝒪𝑐(𝑊, h) lies in such a category 𝒪𝑐,𝑊 ′,𝐿 for some
parabolic subgroup 𝑊 ′ ⊂ 𝑊 and finite dimensional irreducible 𝐿, and the pair (𝑊 ′, 𝐿)
is uniquely determined by 𝑆 up to the natural 𝑊 -action on such pairs.
(4) If (𝑊 ′, 𝐿) labels the simple module 𝑆 in this way, then Supp(𝑆) = 𝑊h𝑊′.
Proof. For the Ext1 statement, notice that the restriction of the parameter 𝑐 to the set
of reflection 𝑆 ′ ⊂ 𝑊 ′ is not only 𝑊 ′-equivariant but also 𝑁𝑊 (𝑊 ′)-equivariant. As the
Euler grading element eu𝑊 ′ is fixed by the 𝑁𝑊 (𝑊 ′)-action, it follows that the lowest
eu𝑊 ′-weight spaces of 𝐿 and 𝐿′ have the same weight, and therefore there can be no
nontrivial extensions between 𝐿 and 𝐿′ by usual highest weight theory (see, e.g. [38,
Section 2]). More precisely, let 𝐿 have lowest weight 𝜆 ∈ Irr(𝑊 ′) and 𝐿′ have lowest
weight 𝜆′ = 𝑛.𝜆 for some 𝑛 ∈ 𝑁𝑊 (𝑊 ′), and consider a module 𝑀 ∈ 𝒪𝑐(𝑊′, h𝑊 ′)
such that [𝑀 ] = [𝐿𝑐(𝜆)] + [𝐿𝑐(𝜆′)] in 𝐾0(𝒪𝑐(𝑊
′, h𝑊 ′)). It suffices to show that
𝑀 ∼= 𝐿𝑐(𝜆) ⊕ 𝐿𝑐(𝜆′). As the lowest weight spaces 𝜆 of 𝐿𝑐(𝜆) and 𝜆′ of 𝐿𝑐(𝜆′) occur
in the same graded degree with respect to the grading by generalized eigenspaces of
eu𝑊 ′ , the lowest weight space of 𝑀 is isomorphic to 𝜆⊕ 𝜆′ as a representation of 𝑊 ′.
It follows that there is an 𝐻𝑐(𝑊′, h𝑊 ′)-module homomorphism ∆𝑐(𝜆) ⊕ ∆𝑐(𝜆
′) →𝑀
that is an isomorphism in the lowest weight space, and as [𝑀 ] = [𝐿𝑐(𝜆)] + [𝐿𝑐(𝜆′)]
this map must annihilate the unique maximal proper submodules of each ∆𝑐(𝜆) and
∆𝑐(𝜆′), as their simple constituents have strictly higher lowest weight spaces, and
hence factor through an isomorphism 𝐿𝑐(𝜆) ⊕ 𝐿𝑐(𝜆′) → 𝑀 , as needed. That 𝒪𝑐,𝑊 ′
and 𝒪𝑐,𝑊 ′,𝐿 are Serre subcategories then follows from the exactness of Res𝑊𝑊 ′ .
Given a simple module 𝑆 ∈ 𝒪𝑐(𝑊, h), its support is of the form 𝑊h𝑊′ for a
parabolic subgroup 𝑊 ′ ⊂ 𝑊 uniquely determined up to conjugation, and Res𝑊𝑊 ′𝑆
is finite dimensional and nonzero for a parabolic subgroup 𝑊 ′ ⊂ 𝑊 if and only if
Supp(𝑆) = 𝑊h𝑊′ , see [5, Proposition 3.2]. So, there exists a finite dimensional
48
simple module 𝐿 ∈ 𝒪𝑐(𝑊′, h𝑊 ′) and a nonzero homomorphism 𝐿 → Res𝑊𝑊 ′𝑆. By
adjunction, there is therefore a nonzero homomorphism Ind𝑊𝑊 ′𝐿 → 𝑆, which is a
surjection because 𝑆 is simple. So it suffices to check Ind𝑊𝑊 ′𝐿 ∈ 𝒪𝑐,𝑊 ′,𝐿, i.e. that the
irreducible constituents of Res𝑊𝑊 ′Ind𝑊𝑊 ′𝐿 lie in the 𝑁𝑊 (𝑊 ′)-orbit of 𝐿. This follows
from the Mackey formula for rational Cherednik algebras at the level of Grothendieck
groups ([66, Lemma 2.5]) and the fact that 𝐿 is annihilated by any nontrivial parabolic
restriction functor.
Remark 3.2.1.3. Proposition 3.2.1.2 can also be obtained from [51, Theorem 3.4.6].
The labeling of simple modules in 𝒪𝑐(𝑊, h) by pairs (𝑊 ′, 𝐿) as in Proposition
3.2.1.2 is an analogue in the setting of rational Cherednik algebras of the partitioning
of irreducible representations of finite groups of Lie type into Harish-Chandra series.
Proposition 3.2.1.4. The image of Ind𝑊𝑊 ′𝐿 in the quotient category 𝒪𝑐,𝑊 ′,𝐿 is a
projective generator. Furthermore, the natural map
End𝒪𝑐,𝑊 ′,𝐿(Ind𝑊𝑊 ′𝐿) → End𝒪𝑐,𝑊 ′,𝐿
(Ind𝑊𝑊 ′𝐿)
is an isomorphism, and in particular there is an equivalence of categories
𝒪𝑐,𝑊 ′,𝐿∼= End𝒪𝑐(𝑊 ′,h𝑊 ′ )(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝-mod𝑓.𝑑..
Proof. The statement comparing the endomorphism algebra of Ind𝑊𝑊 ′𝐿 in 𝒪𝑐,𝑊 ′,𝐿 and
in 𝒪𝑐,𝑊 ′,𝐿 follows from the observation that Ind𝑊𝑊 ′𝐿 has no nonzero submodules or
quotients annihilated by Res𝑊𝑊 ′ and from the definition of morphisms in the quo-
tient category. Otherwise, there would exist a simple module 𝑆 ∈ 𝒪𝑐,𝑊 ′,𝐿 such that
Res𝑊𝑊 ′𝑆 = 0 but such that either Hom𝒪𝑐,𝑊 ′,𝐿(𝑆, Ind𝑊𝑊 ′𝐿) or Hom𝒪𝑐,𝑊 ′,𝐿
(Ind𝑊𝑊 ′𝐿, 𝑆)
is nonzero. As Ind𝑊𝑊 ′ and Res𝑊𝑊 ′ are biadjoint, this is equivalent to one of the hom
spaces Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(Res𝑊𝑊 ′𝑆, 𝐿) or Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,Res𝑊𝑊 ′𝑆) being nonzero, which
contradicts Res𝑊𝑊 ′𝑆 = 0.
The argument in the last paragraph of the proof of Proposition 3.2.1.2 shows that
Ind𝑊𝑊 ′𝐿 admits a surjection to any simple module in 𝒪𝑐,𝑊 ′,𝐿, so the claim that it is
49
a projective generator will follow as soon as we see that it is projective in 𝒪𝑐,𝑊 ′,𝐿.
For this, note that Res𝑊𝑊 ′ and Ind𝑊𝑊 ′ induce a biadjoint pair of functors between
𝒪𝑐,𝑊 ′,𝐿 and the Serre subcategory of 𝒪𝑐(𝑊′, h𝑊 ′) generated by those simple objects
in the 𝑁𝑊 (𝑊 ′)-orbit of 𝐿; indeed, that these functors are biadjoint follows from the
biadjunction of Res𝑊𝑊 ′ and Ind𝑊𝑊 ′ as functors between 𝒪𝑐(𝑊, h) and 𝒪𝑐(𝑊′, h𝑊 ′) and
the facts that every object in 𝒪𝑐,𝑊 ′,𝐿 is of the form 𝜋(𝑀), where 𝜋 : 𝒪𝑐,𝑊 ′,𝐿 → 𝒪𝑐,𝑊 ′,𝐿
is the quotient functor and where 𝑀 ∈ 𝒪𝑐,𝑊 ′,𝐿 is such that all simple objects in
its head and socle have support equal to 𝑊h𝑊′ , that Ind𝑊𝑊 ′𝑀 ′ is such an object
for every 𝑀 ′ in the specified Serre subcategory of 𝒪𝑐(𝑊′, h𝑊 ′), and that 𝜋 is fully
faithful on such objects. This Serre subcategory is semisimple by the Ext1 statement
from Proposition 3.2.1.2, and in particular 𝐿 is projective in that category. Biadjoint
functors preserve projectives, so Ind𝑊𝑊 ′𝐿 is projective in 𝒪𝑐,𝑊 ′,𝐿 as needed.
The category 𝒪𝑐(𝑊, h) is equivalent to the category of finite dimensional modules
of a finite dimensional C-algebra [38, Theorem 5.16]. The same is true for its Serre
subquotient 𝒪𝑐,𝑊 ′,𝐿, which therefore is equivalent to the category of finite dimensional
modules over the opposite endomorphism algebra of any projective generator. The
last statement follows.
Through conjugation and the 𝑊 -action on h, any 𝑤 ∈ 𝑊 determines an isomor-
phism 𝐻𝑐(𝑊′, h𝑊 ′) → 𝐻𝑐(
𝑤𝑊 ′, h𝑤𝑊 ′) and therefore also a bijection
Irr(𝐻𝑐(𝑊′, h𝑊 ′)) → Irr(𝐻𝑐(
𝑤𝑊 ′, h𝑤𝑊 ′)), 𝑀 ↦→ 𝑤𝑀
on the sets of isomorphism classes of irreducible representations by transfer of struc-
ture. The group 𝑊 therefore acts on the set of pairs (𝑊 ′, 𝐿), where 𝑊 ′ ⊂ 𝑊
is a parabolic subgroup and 𝐿 is a finite-dimensional irreducible representation of
𝐻𝑐(𝑊′, h𝑊 ′), by 𝑤.(𝑊 ′, 𝐿) = (𝑤𝑊 ′, 𝑤𝐿). Note that 𝑊 ′ ⊂ 𝑊 stabilizes the pair
(𝑊 ′, 𝐿). This leads to the following definition:
Definition 3.2.1.5. The inertia group 𝐼𝑊 (𝑊 ′, 𝐿) of the pair (𝑊 ′, 𝐿) is the subgroup
𝐼𝑊 (𝑊 ′, 𝐿) := 𝑤 ∈ 𝑊 : 𝑤.(𝑊 ′, 𝐿) = (𝑊 ′, 𝐿)/𝑊 ′ ⊂ 𝑁𝑊 (𝑊 ′)/𝑊 ′.
50
The following statement follows from Propositions 3.2.1.2 and 3.2.1.4:
Corollary 3.2.1.6. The number of isomorphism classes of simple modules in the
category 𝒪𝑐(𝑊, h) labeled by the pair (𝑊 ′, 𝐿) in the sense of Proposition 3.2.1.2 is
the number of irreducible representations of the End𝒪𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝. The number
of isomorphism classes of simple modules in 𝒪𝑐(𝑊, h) with support variety 𝑊h𝑊′ is
given by ∑𝐿∈Irr(𝒪𝑐(𝑊 ′,h𝑊 ′ )),
dimC 𝐿<∞
#Irr(End𝒪𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝)
[𝑁𝑊 (𝑊 ′)/𝑊 ′ : 𝐼𝑊 (𝑊 ′, 𝐿)].
Proof. Immediate from Propositions 3.2.1.2 and 3.2.1.4 and the definition of the in-
ertia group.
Note that Corollary 3.2.1.6 is only non-vacuous in the case 𝑊 ′ = 𝑊 , i.e. for
counting isomorphism classes of simple modules with support strictly containing
𝑊h𝑊 = 0, i.e. for counting isomorphism classes of infinite-dimensional simple
modules in 𝒪𝑐(𝑊, h). But the total number of irreducible representations in 𝒪𝑐(𝑊, h)
is #Irr(𝑊 ), and in this way one also obtains counts of the finite-dimensional simple
modules by subtracting the number of infinite-dimensional simple modules.
The inertia group 𝐼𝑊 ′,𝐿 will be important to our study of End𝒪𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝.
The following is a basic result about the endomorphism algebra:
Proposition 3.2.1.7. We have
dimC End𝐻𝑐(Ind𝑊𝑊 ′𝐿) = #𝐼𝑊 (𝑊 ′, 𝐿)
Proof. By adjunction, we have
End𝒪𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿) ∼= Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,Res𝑊𝑊 ′Ind𝑊𝑊 ′𝐿).
By the Mackey formula for rational Cherednik algebras at the level of Grothendieck
groups [66, Lemma 2.5] and the fact that 𝐿 is finite-dimensional and therefore an-
nihilated by any nontrivial parabolic restriction functor, it follows that the class
of Res𝑊𝑊 ′Ind𝑊𝑊 ′𝐿 in 𝐾0(𝒪𝑐(𝑊′, h𝑊 ′)) equals
∑𝑛∈𝑁𝑊 (𝑊 ′)/𝑊 ′ [𝑛𝐿]. The Ext1 vanishing
51
statement from Proposition 3.2.1.2 therefore implies that Res𝑊𝑊 ′Ind𝑊𝑊 ′𝐿 is isomorphic
to the direct sum ⊕𝑛∈𝑁𝑊 (𝑊 ′)/𝑊 ′𝑛𝐿. We therefore have
Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,Res𝑊𝑊 ′Ind𝑊𝑊 ′𝐿) = Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,⨁
𝑛∈𝑁𝑊 (𝑊 ′)/𝑊 ′
𝑛𝐿).
As Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,𝑛𝐿) is isomorphic to C if 𝑛 ∈ 𝐼𝑊 ′,𝐿 and is 0 otherwise, the claim
follows.
With this dimension result in mind, our goal is to give a presentation of the al-
gebra End𝐻𝑐(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝 as a twisted extension of a Hecke algebra, with a natural
basis indexed by 𝐼𝑊 (𝑊 ′, 𝐿), analogous to the presentation by Howlett and Lehrer
[46] of endomorphism algebras of induced cuspidal representations in the representa-
tion theory of finite groups of Lie type. A first step towards this goal is to realize
End𝐻𝑐(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝 as a quotient of a twist of the group algebra of the fundamen-
tal group of h𝑊 ′𝑟𝑒𝑔/𝐼𝑊 (𝑊 ′, 𝐿) by a group 2-cocycle. This is achieved in the following
section.
3.2.2 The Functor 𝐾𝑍𝐿
Notation: Throughout this section we will again fix a parabolic subgroup 𝑊 ′ ⊂
𝑊 and a finite-dimensional irreducible representation 𝐿 of 𝐻𝑐(𝑊′, h𝑊 ′). The only
parabolic induction and restrictions functors used will be between rational Cherednik
algebras associated to 𝑊 ′ and 𝑊 , so we will omit the superscript 𝑊 and subscript
𝑊 ′ in the notation for the functors Res𝑊𝑊 ′ , Ind𝑊𝑊 ′ , and Res𝑊𝑊 ′ . We will suppress the
basepoint 𝑏 in the notation for the fundamental group of h𝑊 ′𝑟𝑒𝑔 and its quotients, as
this basepoint plays no role in this section.
Lemma 3.2.2.1. The restriction
Res : 𝒪𝑐,𝑊 ′ → (𝒪𝑐(𝑊′, h𝑊 ′) Loc(h𝑊
′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′)
52
of the partial 𝐾𝑍 functor to the subcategory 𝒪𝑐,𝑊 ′ of 𝒪𝑐(𝑊, h) is exact, full, and has
image closed under subquotients. The same is true for its restriction to 𝒪𝑐,𝑊 ′,𝐿.
Proof. Recall that by definition Res is the composition 𝑅𝐻 ∘𝐸 ∘𝜓*∘1h𝑊′
𝑟𝑒𝑔∘𝑊h𝑊
′𝑟𝑒𝑔. All
the functors involved in this composition after 𝑊h𝑊′
𝑟𝑒𝑔are equivalences of categories,
and in particular it suffices to check the claims of the lemma for the functor 𝑊h𝑊′
𝑟𝑒𝑔.
Observe that 𝑊h𝑊′
𝑟𝑒𝑔 is a principal open subset of 𝑊h𝑊′ , defined by the nonvanishing
of the polynomial
𝛿𝑊 ′ :=∑𝑤∈𝑊
𝑤
(∏𝑠/∈𝑆′
𝛼𝑠.
).
Indeed, for any 𝑤,𝑤′ ∈ 𝑊 with 𝑤h𝑊′ = 𝑤′h𝑊
′ the nonvanishing of the polynomial
𝑤(∏
𝑠/∈𝑆′ 𝛼𝑠) defines 𝑤h𝑊 ′𝑟𝑒𝑔 in 𝑤h𝑊
′ and 𝑤′(∏
𝑠/∈𝑆′ 𝛼𝑠) vanishes identically on 𝑤h𝑊′
(𝑤h𝑊 ′ is the common vanishing locus of the 𝑤𝛼𝑠 for 𝑠 ∈ 𝑆 ′, and as 𝑤h𝑊 ′ = 𝑤′h𝑊′
it follows that 𝑆 ′ - and hence its complement in 𝑆 - is not stable under conjuga-
tion by 𝑤−1𝑤′, so 𝑤−1𝑤′(∏
𝑠/∈𝑆′ 𝛼𝑠) vanishes on h𝑊′ and hence 𝑤′(
∏𝑠/∈𝑆′ 𝛼𝑠) vanishes
on 𝑤h𝑊′). Modules 𝑀 in 𝒪𝑐,𝑊 ′ are set-theoretically supported on 𝑊h𝑊
′ , and in
particular the completion functor 𝑊h𝑊′
𝑟𝑒𝑔is simply localization by 𝛿𝑊 ′ .
Define the categories
𝐻𝑐(𝑊, h)-modSupp⊂𝑊h𝑊′ , 𝐻𝑐(𝑊, h)∧𝑊h𝑊
′𝑟𝑒𝑔 -modSupp⊂𝑊h𝑊
′
to be the full subcategories of modules over the respective algebras which are annihi-
lated by sufficiently high powers of the ideal of 𝑊h𝑊′ . The induction functor
𝜃* : 𝐻𝑐(𝑊, h)-modSupp⊂𝑊h𝑊′ → 𝐻𝑐(𝑊, h)∧𝑊h𝑊
′𝑟𝑒𝑔 -modSupp⊂𝑊h𝑊
′
given by extension of scalars along the map of algebras 𝜃 : 𝐻𝑐(𝑊, h) → 𝐻𝑐(𝑊, h)∧𝑊h𝑊′
𝑟𝑒𝑔
amounts to localization by the function 𝛿𝑊 ′ defining 𝑊h𝑊′
𝑟𝑒𝑔 in 𝑊h𝑊′ . We also have
the associated pullback (restriction) functor 𝜃*, and as 𝜃* is localization by a single
element we have 𝜃* ∘ 𝜃* is isomorphic to the identity. It follows that 𝜃* induces an
53
equivalence of categories
𝐻𝑐(𝑊, h)-modSupp⊂𝑊h𝑊′
ker 𝜃*→ 𝐻𝑐(𝑊, h)∧𝑊h𝑊
′𝑟𝑒𝑔 -modSupp⊂𝑊h𝑊
′ .
As 𝒪𝑐,𝑊 ′ is a Serre subcategory of 𝐻𝑐(𝑊, h)-modSupp⊂𝑊h𝑊′ and ker Res = (ker 𝜃*) ∩
𝒪𝑐,𝑊 ′ , the canonical functor
𝒪𝑐,𝑊 ′ →𝐻𝑐(𝑊, h)-modSupp⊂𝑊h𝑊
′
ker 𝜃*
is an inclusion of a full subcategory closed under subquotients, and the result fol-
lows. For identical reasons, the result holds for the restriction of Res to the Serre
subcategories 𝒪𝑐,𝑊 ′,𝐿.
It will be convenient for us to take a semidirect product splitting of the normalizer
𝑁𝑊 (𝑊 ′) as given by the following lemma:
Lemma 3.2.2.2. ([60]) There is a subgroup 𝑁𝑊 ′ ⊂ 𝑁𝑊 (𝑊 ′) such that there is a
semidirect product decomposition 𝑁𝑊 (𝑊 ′) = 𝑊 ′ o𝑁𝑊 ′.
Fix a complement 𝑁𝑊 ′ to 𝑊 ′ in 𝑁𝑊 (𝑊 ′) as in Lemma 3.2.2.2. Let 𝐼𝑊 ′,𝐿 ⊂ 𝑁𝑊 ′
denote the stabilizer of 𝐿 under the action of 𝑁𝑊 ′ on Irr(𝐻𝑐(𝑊′, h𝑊 ′)). The natural
map 𝐼𝑊 ′,𝐿 → 𝐼𝑊 (𝑊 ′, 𝐿) is an isomorphism.
Notation: Throughout this section, in which 𝑊 ′ and 𝐿 are fixed, we will simplify
the notation by denoting 𝐼𝑊 ′,𝐿 by 𝐼.
Given a set ℒ of simple objects in 𝒪𝑐(𝑊′, h𝑊 ′), let ⟨ℒ⟩𝒪𝑐(𝑊 ′,h𝑊 ′ ) denote the Serre
subcategory of 𝒪𝑐(𝑊′, h𝑊 ′) generated by ℒ. Note that the functor Res : 𝒪𝑐,𝑊 ′,𝐿 →
(𝒪𝑐(𝑊′, h𝑊 ′) Loc(h𝑊
′𝑟𝑒𝑔))
𝑁𝑊 (𝑊 ′) factors through the Serre subcategory
(⟨𝑁𝑊 (𝑊 ′).𝐿⟩𝒪𝑐(𝑊 ′,h𝑊 ′ ) Loc(h𝑊
′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′)
⊂(𝒪𝑐(𝑊
′, h𝑊 ′) Loc(h𝑊′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′)
.
Let 𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′) be the quotient
𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′) := 𝐻𝑐(𝑊′, h𝑊 ′)/ ∩𝑛∈𝑁𝑊 (𝑊 ′) Ann𝐻𝑐(𝑊 ′,h𝑊 ′ )(
𝑛𝐿).
54
Note that 𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′) is a semisimple finite-dimensional C-algebra with 𝑁𝑊 (𝑊 ′)-
action and that the category ⟨𝑁𝑊 (𝑊 ′).𝐿⟩𝒪𝑐(𝑊 ′,h𝑊 ′ ) is naturally equivalent to the
category 𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′)-mod𝑓.𝑑.. In particular, choosing a point 𝑏 ∈ h𝑊′
𝑟𝑒𝑔, taking the
fiber of the local system at 𝑏 defines an equivalence of categories
Fiber𝑏 :(⟨𝑁𝑊 (𝑊 ′).𝐿⟩𝒪𝑐(𝑊 ′,h𝑊 ′ ) Loc(h𝑊
′
𝑟𝑒𝑔))𝑁𝑊 (𝑊 ′)
→ 𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′) o 𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝑁𝑊 ′)-mod𝑓.𝑑
where the semidirect product is defined by the natural action of 𝜋1(h𝑊′
𝑟𝑒𝑔/𝑁𝑊 ′) on
𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′) through the natural projection 𝜋1(h𝑊′
𝑟𝑒𝑔/𝑁𝑊 ′) → 𝑁𝑊 ′ . Let 𝑒𝐿 denote the
central idempotent associated to the irreducible representation 𝐿 of 𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′).
Clearly 𝑒𝐿 generates the unit ideal in 𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′)o𝜋1(h𝑊′
𝑟𝑒𝑔/𝑁𝑊 ′) and the associated
spherical subalgebra is given by
𝑒𝐿(𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′) o 𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝑁𝑊 ′))𝑒𝐿 ∼= EndC(𝐿) o 𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼).
In particular, multiplication by 𝑒𝐿 defines an equivalence of categories
𝑒𝐿 : 𝐻𝑐,𝐿(𝑊 ′, h𝑊 ′) o 𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝑁𝑊 ′)-mod𝑓.𝑑 → EndC(𝐿) o 𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)-mod𝑓.𝑑.
The automorphism group of EndC(𝐿) is 𝑃𝐺𝐿C(𝐿), the projective general linear
group of 𝐿, and in particular the action of 𝐼 on End(𝐿) defines a projective represen-
tation 𝜋𝐿 of 𝐼 on 𝐿. Let 𝜇 ∈ 𝑍2(𝐼,C×) be a group 2-cocycle of 𝐼 with coefficients in
C× representing the cohomology class associated to 𝜋𝐿. Let ∈ 𝑍2(𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼),C×)
be the pullback of 𝜇 along the natural projection 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼) → 𝐼. Let − denote
the opposite group cocycle defined by (−)(𝑔) = (𝑔)−1, and for any group 2-
cocycle 𝜈 ∈ 𝑍2(𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼),C×) let C𝜈 [𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)] denote the 𝜈-twisted group algebra
of 𝜋1(h𝑊′
𝑟𝑒𝑔/𝐼), i.e. the associative unital C-algebra with C-basis 𝑒𝑔 : 𝑔 ∈ 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼)
and multiplication 𝑒𝑔𝑒𝑔′ = 𝜈(𝑔, 𝑔′)𝑒𝑔𝑔′ . Essentially by definition of 𝜇 we see that 𝐿 is
55
naturally a C−[𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼)]-module, and in particular we have a functor
HomEndC(𝐿)(𝐿, ∙) : EndC(𝐿) o 𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)-mod𝑓.𝑑. → C−[𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)]-mod𝑓.𝑑..
This functor is an equivalence of categories, with quasi-inverse 𝑁 ↦→ 𝐿⊗C 𝑁 .
Remark 3.2.2.3. Let 𝜋𝐿 : 𝐼 → 𝐺𝐿C(𝐿) be any lift of the projective representation
𝜋𝐿. For any 𝑛 ∈ 𝐼, the operator 𝜋𝐿(𝑛) gives an isomorphism of 𝐻𝑐(𝑊′, h𝑊 ′)-modules
of 𝐿 with its twist 𝑛𝐿 and in particular preserves and is determined by its action
on the lowest eu𝑊 ′-weight space 𝐿0 of 𝐿. It follows that 𝜋𝐿 induces a projective
representation 𝜋𝐿0 of 𝐼 on 𝐿0 with respect to which 𝐿0 is (projectively) equivariant
as a representation of 𝑊 ′. Furthermore, the projective representation 𝜋𝐿 lifts to a
linear representation of 𝐼, and in particular the cocycle 𝜇 is trivial, if and only if
the same holds for the projective representation 𝜋𝐿0 . Indeed, given a representation𝜋𝐿0 of 𝐼 on 𝐿0 with respect to which 𝐿0 is an equivariant representation of 𝑊 ′, this
representation extends uniquely to a representation of 𝐼 on all of 𝐿 with respect to
which 𝐿 is 𝐼-equivariant as an 𝐻𝑐(𝑊′, h𝑊 ′)-module. It follows that triviality of the
cocycle 𝜇 can be checked at the level of the representation of 𝑊 ′ in 𝐿0. For example,
the cocycle 𝜇 is trivial in the large class of examples in which 𝑊 ′ splits as a direct
product 𝑊 ′ = 𝑊 ′1 ×𝑊 ′
2 with respect to which 𝐿0 is isomorphic to a tensor product
𝐿01 ⊗𝐿0
2 and such that both 𝐼 centralizes 𝑊 ′1 and also 𝐿0
2 is the trivial representation.
From the above discussion, we have:
Theorem 3.2.2.4. The Hom functor Hom𝒪𝑐(𝑊,h)(Ind(𝐿), ∙) on 𝒪𝑐,𝑊 ′,𝐿 factors as the
composition of a functor
𝐾𝑍𝐿 : 𝒪𝑐,𝑊 ′,𝐿 → C−[𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)]-mod𝑓.𝑑.
followed by the forgetful functor to the category of finite-dimensional vector spaces.
𝐾𝑍𝐿 is exact, full, has image closed under subquotients, and induces a fully faithful
embedding
𝒪𝑐,𝑊 ′,𝐿 →˓ C−[𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)]-mod𝑓.𝑑.
56
with image closed under subquotients. In particular, there is a surjection of C-algebras
𝜙𝐿 : C−[𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)] → End𝒪𝑐(𝑊,h)(Ind(𝐿))𝑜𝑝𝑝.
Proof. By adjunction, we have Hom𝒪𝑐(𝑊,h)(Ind(𝐿), ∙) ∼= Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,Res∙), and
this latter functor can be expressed as the composition HomEndC(𝐿)(𝐿, ∙)∘𝑒𝐿∘Fiber𝑏∘
Res of functors discussed above, from which the existence of the lift𝐾𝑍𝐿 follows. That
𝐾𝑍𝐿 is exact, full, and has image closed under subquotients follows from Lemma
3.2.2.1 and the fact that the functors appearing after Res in the composition above
are equivalences of categories. That the induced functor on 𝒪𝑐,𝑊 ′,𝐿 is faithful follows
from the fact that it is represented by the projective generator Ind(𝐿). By Proposition
3.2.1.4, this induced functor is identified with a fully faithful embedding
End𝒪𝑐(𝑊,h)(Ind(𝐿))𝑜𝑝𝑝-mod𝑓.𝑑. → C−[𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)]-mod𝑓.𝑑.
with image closed under subquotients and which is identity at the level of C-vector
spaces. Such a functor is simply restriction along some algebra homomorphism
𝜙𝐿 : C−[𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼)] → End𝒪𝑐(𝑊,h)(Ind(𝐿))𝑜𝑝𝑝. The image im 𝜙𝐿 is a C−[𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)]-
submodule of End𝒪𝑐(𝑊,h)(Ind(𝐿))𝑜𝑝𝑝, and because the functor in question has im-
age closed under subquotients it follows that im 𝜙𝐿 is also a End𝒪𝑐(𝑊,h)(Ind(𝐿))𝑜𝑝𝑝-
submodule. As im 𝜙𝐿 contains 1, it follows that 𝜙𝐿 is surjective, as needed.
Definition 3.2.2.5. Let the generalized Hecke algebra ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) attached to
𝑐,𝑊 ′, 𝐿 and the group 𝑊 be the quotient algebra
ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) := C−[𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)]/ ker𝜙𝐿.
Let
𝐾𝑍𝐿 : 𝒪𝑐,𝑊 ′,𝐿 → ℋ(𝑐,𝑊 ′, 𝐿,𝑊 )-mod𝑓.𝑑.
be the natural factorization of the functor 𝐾𝑍𝐿 from Theorem 3.2.2.4 through the
category ℋ(𝑐,𝑊 ′, 𝐿,𝑊 )-mod𝑓.𝑑..
57
Corollary 3.2.2.6. 𝐾𝑍𝐿 induces an equivalence of categories
𝒪𝑐,𝑊 ′,𝐿∼= ℋ(𝑐,𝑊 ′, 𝐿,𝑊 )-mod𝑓.𝑑..
Remark 3.2.2.7. In the case 𝑊 ′ = 1 and 𝐿 = C, the generalized Hecke algebra
ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) is precisely the Hecke algebra H𝑞(𝑊 ) attached to the complex reflection
group 𝑊 , the functor 𝐾𝑍C is precisely the 𝐾𝑍 functor of [38], and the statements of
Theorem 3.2.2.4 are well-known in that case. The equivalence in Corollary 3.2.2.6 in
that case is the well-known equivalence 𝒪𝑐/𝒪𝑡𝑜𝑟𝑐
∼= H𝑞(𝑊 )-mod𝑓.𝑑..
The following remark reduces the study of the algebras ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) to the case
in which 𝑊 is an irreducible complex reflection group.
Remark 3.2.2.8. Suppose the complex reflection group 𝑊 and its reflection repre-
sentation decompose as a product (𝑊, h) = (𝑊1 × 𝑊2, h𝑊1 ⊕ h𝑊2). Let 𝑆𝑖 ⊂ 𝑊𝑖
be the set of reflections in 𝑊𝑖 for 𝑖 = 1, 2, so that 𝑆 = 𝑆1 ⊔ 𝑆2, and let 𝑐𝑖 be the
restriction of the parameter 𝑐 to 𝑆𝑖. The rational Cherednik algebra 𝐻𝑐(𝑊, h) de-
composes naturally as the tensor product 𝐻𝑐1(𝑊1, h𝑊1) ⊗𝐻𝑐2(𝑊2, h𝑊2). Let 𝑊 ′ ⊂ 𝑊
be a parabolic subgroup, and for 𝑖 = 1, 2 let 𝑊 ′𝑖 ⊂ 𝑊𝑖 be the parabolic subgroups
such that 𝑊 ′ = 𝑊 ′1 ×𝑊 ′
2. Let 𝐿𝑖 be a finite-dimensional irreducible representation
of 𝐻𝑐𝑖(𝑊′𝑖 , h𝑊 ′
𝑖) for 𝑖 = 1, 2, so that 𝐿 = 𝐿1 ⊗ 𝐿2 is a finite-dimensional irreducible
representation of 𝐻𝑐(𝑊′1×𝑊 ′
2, h𝑊 ′1×𝑊 ′
2) = 𝐻𝑐1(𝑊
′1, h𝑊 ′
1)⊗𝐻𝑐2(𝑊
′2, h𝑊 ′
2). Every finite-
dimensional irreducible representation of 𝐻𝑐(𝑊′1×𝑊 ′
2, h𝑊 ′1×h𝑊 ′
2) appears in this way,
and all of the constructions appearing in Theorem 3.2.2.4 split naturally as well. In
particular, there is a natural isomorphism of algebras
ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) ∼= ℋ(𝑐1,𝑊′1, 𝐿1,𝑊1) ⊗ℋ(𝑐2,𝑊
′2, 𝐿2,𝑊2)
compatible with the surjections 𝜙𝐿1 , 𝜙𝐿2 and 𝜙𝐿1⊗𝐿2.
58
3.2.3 Eigenvalues of Monodromy and Relations From Corank
1
We want to describe the generalized Hecke algebras ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) as explicitly as
possible so that we can in turn understand the subquotient categories 𝒪𝑐,𝑊 ′,𝐿. To
achieve this, we study the kernel ker𝜙𝐿.
Let 𝑊 ′′ ⊂ 𝑊 be another parabolic subgroup, with 𝑊 ′ ⊂ 𝑊 ′′ ⊂ 𝑊 . Let
𝐼𝑊 ′,𝐿,𝑊 ′′ := 𝐼𝑊 ′,𝐿∩𝑊 ′′ be the (lift to 𝑁𝑊 ′′(𝑊 ) via the splitting 𝑁𝑊 (𝑊 ′) = 𝑊 ′o𝑁𝑊 ′
of the) inertia group of 𝐿 in 𝑁𝑊 ′′(𝑊 ′). The 2-cocycle 𝜇 ∈ 𝑍2(𝐼𝑊 ′,𝐿,C×) restricts
to give a 2-cocycle of 𝐼𝑊 ′,𝐿,𝑊 ′′ , which we will also denote by 𝜇, and by pullback de-
termines a 2-cocycle of 𝜋1((h𝑊 ′′)𝑊′
𝑟𝑒𝑔/𝐼𝑊 ′,𝐿,𝑊 ′′), which we will also denote by . We
may choose 𝑏 ∈ h𝑊′
𝑟𝑒𝑔 such that the projection 𝑏𝑊 ′′ of 𝑏 to (h𝑊 ′′)𝑊′ with respect to
the vector space decomposition h𝑊′
= (h𝑊 ′′)𝑊′ ⊕ h𝑊
′′ lies in (h𝑊 ′′)𝑊′
𝑟𝑒𝑔. By Theorem
3.2.2.4, there is a surjection of C-algebras
𝜙𝐿,𝑊 ′′ : C−[𝜋1((h𝑊 ′′)𝑊′
𝑟𝑒𝑔/𝐼𝑊 ′,𝐿,𝑊 ′′ , 𝑏𝑊 ′′)] → End𝒪𝑐(𝑊 ′′,h𝑊 ′′ )(Ind𝑊′′
𝑊 ′ 𝐿)𝑜𝑝𝑝.
Notation: Throughout this section, the parabolic subgroup 𝑊 ′ and finite dimen-
sional irreducible representation 𝐿 of 𝐻𝑐(𝑊′, h𝑊 ′) will remain fixed, and the parabolic
subgroup 𝑊 ′′ will vary. We will denote 𝐼𝑊 ′,𝐿 by 𝐼, as in the previous section, and we
will denote 𝐼𝑊 ′,𝐿,𝑊 ′′ by 𝐼𝑊 ′′ . We will always take the basepoints for the fundamental
groups of h𝑊 ′𝑟𝑒𝑔 and (h𝑊 ′′)𝑊
′𝑟𝑒𝑔 and their quotients to be 𝑏 and 𝑏𝑊 ′′ as above, and we will
suppress these from the notation for readability.
By the construction of the functor 𝜄*𝑊 ′′,𝑊 ′ of Gordon and Martino recalled in
Section 2.6 and the fact that 𝐼𝑊 ′′ acts trivially on h𝑊′′
𝑟𝑒𝑔 , there is a natural map of
fundamental groups
𝜄𝑊 ′′,𝑊 ′,𝐿 : 𝜋1((h𝑊 ′′)𝑊′
𝑟𝑒𝑔/𝐼𝑊 ′′) → 𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼),
59
extending to a natural map of C-algebras
𝜄𝑊 ′′,𝑊 ′,𝐿 : C−[𝜋1((h𝑊 ′′)𝑊′
𝑟𝑒𝑔/𝐼𝑊 ′′)] → C−[𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)].
By functoriality, there is also an algebra homomorphism
End𝒪𝑐(𝑊 ′′,h𝑊 ′′ )(Ind𝑊′′
𝑊 ′ 𝐿)𝑜𝑝𝑝 → End𝒪𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝
induced by the functor Ind𝑊𝑊 ′′ and the isomorphism Ind𝑊𝑊 ′′ ∘ Ind𝑊′′
𝑊 ′ ∼= Ind𝑊𝑊 ′ . The
following lemma is then immediate from the constructions and from Gordon and
Martino’s transitivity result that was recalled in Theorem 2.6.0.1:
Lemma 3.2.3.1. The following diagram commutes:
C−[𝜋1((h𝑊 ′′)𝑊′
𝑟𝑒𝑔/𝐼𝑊 ′′)]𝜄𝑊 ′′,𝑊 ′,𝐿
> C−[𝜋1(h𝑊 ′
𝑟𝑒𝑔/𝐼)]
End𝒪𝑐(𝑊 ′′,h𝑊 ′′ )(Ind𝑊′′
𝑊 ′ 𝐿)𝑜𝑝𝑝
𝜙𝐿,𝑊 ′′∨Ind𝑊
𝑊 ′′> End𝒪𝑐(𝑊,h)(Ind𝑊𝑊 ′𝐿)𝑜𝑝𝑝
𝜙𝐿∨
Definition 3.2.3.2. Given a chain of parabolic subgroups 𝑊 ′ ⊂ 𝑊 ′′ ⊂ 𝑊 as above,
we say that 𝑊 ′ is of corank 𝑟 in 𝑊 ′′ if dim h𝑊 ′′ = dim h𝑊 ′ + 𝑟.
The parabolic subgroups 𝑊 ′′ ⊂ 𝑊 containing 𝑊 ′ in corank 1 are closely related
to the reflections appearing in the action of 𝑁𝑊 (𝑊 ′) on h𝑊′ :
Lemma 3.2.3.3. Let 𝑊 ′′ ⊂ 𝑊 be a parabolic subgroup containing 𝑊 ′ in corank 1.
The quotient 𝑁𝑊 ′′(𝑊 ′)/𝑊 ′ is cyclic and acts on h𝑊′ by complex reflections through
the hyperplane h𝑊′′ ⊂ h𝑊
′. Furthermore, every element 𝑛 ∈ 𝑁𝑊 (𝑊 ′) that acts on
h𝑊′ as a complex reflection lies in some parabolic subgroup 𝑊 ′′ ⊂ 𝑊 containing 𝑊 ′
in corank 1.
Proof. The action of 𝑁𝑊 ′′(𝑊 ′)/𝑊 ′ on h𝑊′ is faithful because 𝑊 ′ is precisely the
pointwise stabilizer of h𝑊 ′ in 𝑊 , and the action respects the decomposition h𝑊′
=
(h𝑊 ′′)𝑊′ ⊕h𝑊
′′ . As 𝑁𝑊 ′′(𝑊 ′)/𝑊 ′ acts trivially on h𝑊′′ and dim(h𝑊 ′′)𝑊
′= 1, the first
claim follows.
60
For the second claim, suppose 𝑛 ∈ 𝑁𝑊 (𝑊 ′) acts on h𝑊′ as a complex reflection
through the hyperplane 𝐻 ⊂ h𝑊′ . Let 𝑊 ′′ ⊂ 𝑊 be the point-wise stabilizer of
𝐻 in 𝑊 , a parabolic subgroup. As 𝑛 does not fix h𝑊′ but 𝑊 ′ ⊂ 𝑊 ′′, we have
𝐻 ⊂ h𝑊′′ ( h𝑊
′ . As 𝐻 is a hyperplane, it follows that 𝐻 = h𝑊′′ , so 𝑊 ′′ is a
parabolic subgroup containing 𝑊 ′ in corank 1 and 𝑛 ∈ 𝑁𝑊 ′′(𝑊 ) acts on h𝑊′ as a
complex reflection through the hyperplane h𝑊′′ ⊂ h𝑊
′ , as needed.
It follows from Lemma 3.2.3.3 that the inertia group 𝐼𝑊 ′′ considered above is
cyclic and acts on (h𝑊 ′′)𝑊′
𝑟𝑒𝑔 through a faithful character. As 𝑊 ′ is corank 1 in 𝑊 ′′,
we have (h𝑊 ′′)𝑊′
𝑟𝑒𝑔∼= C× as a C-manifold, and in particular there is a canonical group
isomorphism
𝜋1((h𝑊 ′′)𝑊′
𝑟𝑒𝑔/𝐼𝑊 ′′) ∼= Z.
Definition 3.2.3.4. Let 𝑇𝑊 ′,𝐿,𝑊 ′′ denote the canonical generator of 𝜋1((h𝑊 ′′)𝑊′
𝑟𝑒𝑔/𝐼𝑊 ′′)
arising from the isomorphism above. We will also let 𝑇𝑊 ′,𝐿,𝑊 ′′ denote its image in
𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼) under the homomorphism 𝜄𝑊 ′′,𝑊 ′,𝐿, and we refer to 𝑇𝑊 ′,𝐿,𝑊 ′′ as the gener-
ator of monodromy about the hyperplane h𝑊′′ ⊂ h𝑊
′. We call the parabolic subgroup
𝑊 ′′ and its associated hyperplane h𝑊′′ ⊂ h𝑊
′𝐿-trivial (resp., 𝐿-essential) when the
inertia group 𝐼𝑊 ′′ is trivial (resp., nontrivial). Let h𝑊 ′𝐿−𝑟𝑒𝑔 denote the complement of
the arrangement of 𝐿-essential hyperplanes in h𝑊′, and let 𝐼𝑟𝑒𝑓 denote the subgroup
of 𝐼 generated by the subgroups 𝐼𝑊 ′′ for all 𝐿-essential 𝑊 ′′.
We comment that 𝐼𝑟𝑒𝑓 is a normal subgroup of 𝐼 and that the action of 𝐼𝑟𝑒𝑓 on
h𝑊′ is generated by complex reflections through the 𝐿-essential hyperplanes. In fact,
from Lemma 3.2.3.3 we see that 𝐼𝑟𝑒𝑓 is the maximal reflection subgroup of 𝐼 with
respect to its representation in h𝑊′ . We have h𝑊 ′
𝑟𝑒𝑔 ⊂ h𝑊′
𝐿−𝑟𝑒𝑔, and h𝑊′
𝐿−𝑟𝑒𝑔 is stable under
the action of 𝐼 on h𝑊′ .
The second cohomology group 𝐻2(𝐶,C×) vanishes for all cyclic groups 𝐶, and
in particular we may assume that the 2-cocycle 𝜇 on 𝐼 is trivial on all subgroups
𝐼𝑊 ′′ ⊂ 𝐼 associated to parabolic subgroups 𝑊 ′′ ⊂ 𝑊 containing 𝑊 ′ in corank 1.
Then, the twisted group algebra C−[𝜋1((h𝑊 ′′)𝑊′
𝑟𝑒𝑔/𝐼𝑊 ′′)] is naturally identified with
the Laurent polynomial ring C[𝑇±1𝑊 ′,𝐿,𝑊 ′′ ]. The kernel of the map 𝜙𝐿,𝑊 ′′ appearing
61
in the commutative diagram in Lemma 3.2.3.1 is therefore generated by a monic
polynomial 𝑃𝑊 ′,𝐿,𝑊 ′′ ∈ C[𝑇𝑊 ′,𝐿,𝑊 ′′ ] nonvanishing at 0. It follows from Proposition
3.2.1.7 that 𝑃𝑊 ′,𝐿,𝑊 ′′ is a polynomial of degree #𝐼𝑊 ′′ .
Definition 3.2.3.5. Let 𝑃𝑊 ′,𝐿,𝑊 ′′ ∈ C[𝑇 ] be the monic polynomial of degree #𝐼𝑊 ′′
generating the kernel ker𝜙𝑊 ′,𝐿,𝑊 ′′.
Note that 𝑃𝑊 ′,𝐿,𝑊 ′′ only depends on 𝑊 ′′ up to 𝐼-conjugacy and that 𝑃𝑊 ′,𝐿,𝑊 ′′(0) =
0.
Notation: As 𝑊 ′ and 𝐿 remain fixed, we will denote 𝑃𝑊 ′,𝐿,𝑊 ′′ by 𝑃𝑊 ′′ and 𝑇𝑊 ′,𝐿,𝑊 ′′
by 𝑇𝑊 ′′ when the meaning is clear.
We can now state the main result of this section.
Theorem 3.2.3.6. The map 𝜙𝐿 of Theorem 3.2.2.4 factors through a surjection
𝜙𝐿 :C−[𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩→ ℋ(𝑐,𝑊 ′, 𝐿,𝑊 )
of C-algebras. If the inequality
dimC−[𝜋1(h
𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′′ is 𝐿-essential⟩≤ #𝐼𝑟𝑒𝑓
holds then it is an equality and 𝜙𝐿 is an isomorphism.
In particular, if 𝑊 is a Coxeter group or if the group 2-cocycle 𝜇 ∈ 𝑍2(𝐼,C×) has
cohomologically trivial restriction to 𝐼𝑟𝑒𝑓 , 𝜙𝐿 is an isomorphism.
Remark 3.2.3.7. Note that only the 𝐿-essential hyperplanes feature in dimension
bound above. The 2-cocycle ∈ 𝑍2(𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 ),C×) is obtained by pulling back
the 2-cocycle 𝜇 ∈ 𝑍2(𝐼,C×) along the natural projection. Note that by their definition,
the generators of monodromy 𝑇𝑊 ′′ are naturally elements of 𝜋1(h𝑊′
𝑟𝑒𝑔/𝐼𝑟𝑒𝑓 ).
The following lemma will be useful in the proof of Theorem 3.2.3.6
62
Lemma 3.2.3.8. There is a linear character 𝜒 : 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼) → C× satisfying
𝜒(𝑇𝑊 ′′) =
⎧⎪⎨⎪⎩−𝑃𝑊 ′′(0) if 𝑊 ′′ is 𝐿-trivial
1 otherwise.
Proof. As 𝑃𝑊 ′′(0) is nonzero and only depends on 𝑊 ′′ up to 𝐼-conjugacy, it suffices
to show that for any parabolic subgroup 𝑊 ′′ containing 𝑊 ′ in corank 1 there exists
a group homomorphism 𝜃 : 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼) → Z such that a generator of monodromy 𝑇𝐻
about one of the hyperplanes 𝐻 defining h𝑊′
𝑟𝑒𝑔/𝐼 is sent to 1 under 𝜃 if and only if
𝐻 is 𝐼-conjugate to h𝑊′′ and 0 otherwise. One may then compose with the group
homomorphism Z → C×, 1 ↦→ −𝑃𝑊 ′′(0), and take the product of such maps over
all 𝐼-conjugacy classes of 𝐿-trivial parabolic subgroups 𝑊 ′′ ⊃ 𝑊 ′. To construct
such a homomorphism, choose a linear functional 𝛼 ∈ (h𝑊′)* defining h𝑊
′′ in h𝑊′
and let 𝛿 =∏
𝑛∈𝐼 𝑛𝛼. Then, 𝛿 defines a continuous function 𝛿 : h𝑊′
𝑟𝑒𝑔/𝐼 → C× with
associated map 𝜋1(𝛿) : 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼) → 𝜋1(C×) = Z. That 𝜋1(𝛿) has the desired effect
on the generators of monodromy then follows easily as in the proof of [7, Proposition
2.16].
Proof of Theorem 3.2.3.6. That 𝜙𝐿 factors as 𝜙𝐿 through the quotient above follows
immediately from Lemma 3.2.3.1.
Assume the dimension inequality in the theorem statement holds. As the action of
𝐼𝑟𝑒𝑓 on h𝑊′ is generated by reflections, the quotient h𝑊 ′
/𝐼𝑟𝑒𝑓 is a smooth C-variety by
the Chevalley-Shephard-Todd theorem [11, 69]. In particular, by Proposition A.1 of
[7] and induction it follows that the kernel of the map 𝜋1(h𝑊′
𝑟𝑒𝑔/𝐼𝑟𝑒𝑓 ) → 𝜋1(h
𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )
is generated (as a normal subgroup) by the generators of monodromy 𝑇𝑊 ′′ for the 𝐿-
trivial 𝑊 ′′. In particular, we have an isomorphism
𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )
⟨𝑇𝑊 ′′ : 𝑊 ′′ is 𝐿-trivial⟩∼= 𝜋1(h
𝑊 ′
𝐿−𝑟𝑒𝑔/𝐼𝑟𝑒𝑓 ).
Clearly, this isomorphism respects the cocycle . Let 𝜒 : 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 ) → C× be the
composition of the natural map 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 ) → 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼) with the character from
63
Lemma 3.2.3.8. The assignments 𝑔 ↦→ 𝜒(𝑔)𝑔 for 𝑔 ∈ 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼) extend to an algebra
automorphism 𝜏𝜒 of C−[𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼)] satisfying
𝜏𝜒(𝑃𝑊 ′′(𝑇𝑊 ′′)) =
⎧⎪⎨⎪⎩−𝑃𝑊 ′′(0)(𝑇𝑊 ′′ − 1) if 𝑊 ′′ is 𝐿-trivial
𝑃𝑊 ′′(𝑇𝑊 ′′) otherwise..
To see this formula in the case that 𝑊 ′′ is 𝐿-trivial, recall that in that case 𝑃𝑊 ′′(𝑇𝑊 ′′)
is a monic linear polynomial, hence of the form 𝑃𝑊 ′′(𝑇𝑊 ′′) = 𝑇𝑊 ′′ +𝑃𝑊 ′′(0); applying
𝜏𝜒 gives
𝜏𝜒(𝑃𝑊 ′′(𝑇𝑊 ′′)) = 𝜏𝜒(𝑇𝑊 ′′ + 𝑃𝑊 ′′(0))
= −𝑃𝑊 ′′(0)𝑇𝑊 ′′ + 𝑃𝑊 ′′(0) = −𝑃𝑊 ′′(0)(𝑇𝑊 ′′ − 1),
as above. The formula in the case that 𝑊 ′′ is 𝐿-essential is clear. Composing these
two isomorphisms yields an isomorphism
C−[𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′′ is 𝐿-essential⟩∼=
C−[𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩.
By assumption, this algebra has dimension at most #𝐼𝑟𝑒𝑓 , and so the same holds for
the dimension 𝑑 ≤ #𝐼𝑟𝑒𝑓 of its image in
C−[𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩
under the natural map induced by the map 𝜋1(h𝑊′
𝑟𝑒𝑔/𝐼𝑟𝑒𝑓 ) → 𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼) of fundamental
groups. As 𝜋1(h𝑊′
𝑟𝑒𝑔/𝐼)/𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 ) ∼= 𝐼/𝐼𝑟𝑒𝑓 , it follows that
dimC−[𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩≤ 𝑑 · (#𝐼/𝐼𝑟𝑒𝑓 ) ≤ #𝐼.
But, as 𝜙𝐿 is a surjection and dimℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) = #𝐼 by Proposition 3.2.1.7, it
follows also that
#𝐼 ≤ dimC−[𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩.
64
It follows that the dimension inequalities above are equalities and that 𝜙𝐿 is an
isomorphism.
For the final statement of the theorem, first suppose the restriction of 𝜇 to 𝐼𝑟𝑒𝑓 is
cohomologically trivial. Then the quotient
C[𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′′ is 𝐿-essential⟩
is precisely a specialization to C of the Hecke algebra attached to the complex re-
flection group 𝐼𝑟𝑒𝑓 in the sense of Broué-Malle-Rouquier [7]. In that paper it was
conjectured that the generic Hecke algebra is a free module of rank #𝐼𝑟𝑒𝑓 over its
ring of parameters, and this conjecture was subsequently proved in characteristic zero
[27]. In particular, this algebra has dimension #𝐼𝑟𝑒𝑓 , as needed.
Now suppose 𝑊 is a Coxeter group. In that case, 𝐼𝑟𝑒𝑓 is a Coxeter group as well,
and its action on h𝑊′ is the complexification of a real reflection representation. We
need to show that the algebra
C−[𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′′ is 𝐿-essential⟩
has dimension at most, and hence equal to, #𝐼𝑟𝑒𝑓 . In this case, the fundamental
group 𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 ) is the Artin braid group 𝐵𝐼𝑟𝑒𝑓 attached to the Coxeter group
𝐼𝑟𝑒𝑓 . If 𝑆 ⊂ 𝐼𝑟𝑒𝑓 is a set of simple reflections for this Coxeter group, the group 𝐵𝐼𝑟𝑒𝑓
is generated by the generators of monodromy 𝑒𝑠 : 𝑠 ∈ 𝑆 about the hyperplanes
associated to the simple reflections, with braid relations as recalled in Section 2.7.
Given a finite list s = (s1, ..., sn) of simple reflections, let 𝑒s denote the product
𝑒𝑠1 · · · 𝑒𝑠𝑛 . By Matsumoto’s Theorem (see [37, Theorem 1.2.2]), for any 𝑤 ∈ 𝐼𝑟𝑒𝑓
the element 𝑒s does not depend on the choice of reduced expression 𝑤 = 𝑠1 . . . 𝑠𝑛,
and we denote as usual the element 𝑒s by 𝑒𝑤 in this case. For a list s let 𝑇s denote
the element 𝑒s viewed as an element of the quotient algebra above, and similarly
for 𝑇𝑤. To show that the algebra has dimension at most #𝐼𝑟𝑒𝑓 , it suffices to check
that the C-span of the elements 𝑇𝑤 : 𝑤 ∈ 𝐼𝑟𝑒𝑓 is closed under multiplication by
65
𝑇𝑠 for any 𝑠 ∈ 𝑆. If 𝑠𝑠1 · · · 𝑠𝑛 is a reduced expression with 𝑤 = 𝑠1 · · · 𝑠𝑛 a reduced
expression for 𝑤, then 𝑇𝑠𝑇𝑤 = 𝜇(𝑠, 𝑤)−1𝑇𝑠𝑤. Otherwise, we may choose a reduced
expression 𝑤 = 𝑠1 · · · 𝑠𝑛 for 𝑤 with 𝑠1 = 𝑠. The hyperplane associated to 𝑠 is 𝐿-
essential and #𝐼𝑊 ′′ = 2 for all 𝐿-essential 𝑊 ′′ in this case, so 𝑇𝑠 satisfies a quadratic
relation 𝑇 2𝑠 = 𝑎+ 𝑏𝑇𝑠. We then have 𝑇𝑠𝑇𝑤 = 𝜇(𝑠, 𝑠𝑤)𝑇 2
𝑠 𝑇𝑠𝑤 = 𝜇(𝑠, 𝑠𝑤)(𝑎+ 𝑏𝑒𝑠)𝑇𝑠𝑤 =
𝜇(𝑠, 𝑠𝑤)(𝑎𝑇𝑠𝑤 +𝜇(𝑠, 𝑠𝑤)−1𝑏𝑇𝑤) = 𝜇(𝑠, 𝑠𝑤)𝑎𝑇𝑠𝑤 + 𝑏𝑇𝑤, which lies in the desired space,
as needed.
3.3 The Coxeter Group Case
In this section, suppose 𝑊 is a finite Coxeter group with set of simple reflections 𝑆 ⊂
𝑊 and real reflection representation hR with inner product ⟨·, ·⟩. Let h := C⊗R hR be
the complexified reflection representation, let 𝑐 : 𝑆 → C be a𝑊 -invariant function (by
which we mean 𝑐(𝑠) = 𝑐(𝑠′) whenever 𝑠 and 𝑠′ are conjugate in 𝑊 ), and let 𝐻𝑐(𝑊, h)
be the associated rational Cherednik algebra (note that 𝑐 extends uniquely to a 𝑊 -
invariant function on the set of reflections in 𝑊 ). We will use certain simplifications
that arise in the Coxeter case, such as natural splittings of the normalizers of parabolic
subgroups and the 𝑇𝑤-basis of the Hecke algebra H𝑞(𝑊 ), to significantly improve
upon the description of the generalized Hecke algebras ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ′′) studied in the
previous section. In particular, we will give a natural presentation for these algebras
and will explain how to compute the quadratic relations 𝑃𝑊 ′,𝐿,𝑊 ′′(𝑇𝑊 ′,𝐿,𝑊 ′′) = 0 for
the 𝐿-essential parabolic subgroups 𝑊 ′′ ⊃ 𝑊 ′ introduced in Definition 3.2.3.5.
Notation: As in previous sections, we will take𝑊 ′ and 𝐿 to be fixed and will suppress
them from the notation, e.g. write 𝐼𝑊 ′′ rather than 𝐼𝑊 ′,𝐿, when the meaning is clear.
3.3.1 A Presentation of the Endomorphism Algebra
Let 𝐽 ⊂ 𝑆 be a subset of the simple reflections and let 𝑊 ′ := ⟨𝐽⟩ be the parabolic
subgroup of 𝑊 generated by 𝐽 . Let 𝐻𝑐(𝑊′, h𝑊 ′) be the associated rational Cherednik
algebra as considered in previous sections, and let 𝐿 be a finite-dimensional irreducible
representation of 𝐻𝑐(𝑊′, h𝑊 ′). Let h𝑊 ′,R ⊂ hR denote the unique 𝑊 ′-stable comple-
66
ment to h𝑊′
R in hR. We have h𝑊′
= (h𝑊′
R )C and h𝑊 ′ = (h𝑊 ′,R)C, where the subscript
C denotes the complexification viewed as a subspace of h. The action of 𝑊 ′ on h𝑊 ′,R
is naturally identified with the real reflection representation of the Coxeter system
(𝑊 ′, 𝐽). Let h𝑊′
R,𝑟𝑒𝑔 ⊂ h𝑊′
R denote the subset of points with stabilizer in 𝑊 equal to
𝑊 ′. The complexification (h𝑊′
R,𝑟𝑒𝑔)C is a proper subspace of h𝑊 ′𝑟𝑒𝑔.
Let Φ ⊂ hR denote the root system attached to (𝑊,𝑆), let Φ+ ⊂ Φ denote the
set of positive roots, and for 𝑠 ∈ 𝑆 let 𝛼𝑠 ∈ Φ denote the positive simple root
defining the reflection 𝑠. For a subset 𝐽 ⊂ 𝑆 of the simple roots, let Φ𝐽 ⊂ Φ denote
the associated root subsystem with positive roots Φ+𝐽 = Φ𝐽 ∩ Φ+. Let 𝒞𝐽 := 𝑥 ∈
h𝑊 ′,R : ⟨𝛼𝑠, 𝑥⟩ > 0 for all 𝑠 ∈ 𝐽 ⊂ h𝑊 ′,R be the associated open fundamental Weyl
chamber. In [47, Corollary 3], Howlett explains that the normalizer 𝑁𝑊 (𝑊 ′) splits
as a semidirect product 𝑁𝑊 (𝑊 ′) = 𝑊 ′ o 𝑁𝑊 ′ , where 𝑁𝑊 ′ is the set-wise stabilizer
of 𝒞𝐽 in 𝑁𝑊 (𝑊 ′). We then take the inertia subgroup 𝐼 ⊂ 𝑁𝑊 ′ to be the stabilizer
in 𝑁𝑊 ′ of the representation 𝐿, as defined after Lemma 3.2.2.2. Recall that for
each parabolic subgroup 𝑊 ′′ ⊂ 𝑊 containing 𝑊 ′ in corank 1 we have the associated
subgroup 𝐼𝑊 ′′ := 𝐼∩𝑊 ′′, and recall that 𝐼𝑟𝑒𝑓E𝐼 is the normal subgroup generated by
the 𝐼𝑊 ′′ for all such 𝑊 ′′. The subgroups 𝐼𝑊 ′′ are all either trivial (for 𝐿-trivial 𝑊 ′′) or
of order 2 (for 𝐿-essential 𝑊 ′′) acting on h𝑊′
R through orthogonal reflection through
the hyperplane h𝑊′′
R ⊂ h𝑊′
R . Therefore 𝐼𝑟𝑒𝑓 is a real reflection group with faithful
reflection representation h𝑊′
R (potentially with nontrivial fixed points). Recall that
𝐼𝑟𝑒𝑓 is the maximal reflection subgroup of 𝐼.
The complement h𝑊 ′
R,𝐿−𝑟𝑒𝑔 of the real hyperplanes h𝑊 ′′
R ⊂ h𝑊′
R is the locus of points
in h𝑊′
R with trivial stabilizer in 𝐼𝑟𝑒𝑓 . By the standard theory of real reflection groups,
𝐼𝑟𝑒𝑓 acts simply transitively on the set of connected components of h𝑊 ′
R,𝐿−𝑟𝑒𝑔. Choose
a connected component 𝒞 ⊂ h𝑊′
R,𝐿−𝑟𝑒𝑔, and let 𝐼𝑐𝑜𝑚𝑝 ⊂ 𝐼 be the set-wise stabilizer of 𝒞
in 𝐼. Clearly, we then have the semidirect product decomposition
𝐼 = 𝐼𝑟𝑒𝑓 o 𝐼𝑐𝑜𝑚𝑝.
Choosing the connected component 𝒞 amounts to choosing a fundamental Weyl cham-
67
ber for 𝐼𝑟𝑒𝑓 , and in particular the pair (𝐼𝑟𝑒𝑓 , 𝑆𝑊 ′,𝐿) is a Coxeter system, where 𝑆𝑊 ′,𝐿
is the set of reflections through the walls of 𝒞. As 𝐼𝑐𝑜𝑚𝑝 acts on 𝒞, it follows that
𝐼𝑐𝑜𝑚𝑝 permutes the walls of 𝐶, and in particular the action of 𝐼𝑐𝑜𝑚𝑝 on 𝐼𝑟𝑒𝑓 is through
diagram automorphisms of the Dynkin diagram of (𝐼𝑟𝑒𝑓 , 𝑆𝑊 ′,𝐿).
Let 𝑞 : 𝑆𝑊 ′,𝐿 → C×, 𝑠 ↦→ 𝑞𝑠, be the unique 𝐼-invariant function such that for
each 𝑠 ∈ 𝑆𝑊 ′,𝐿 the quadratic polynomial 𝑃𝑊 ′,𝐿,⟨𝑊 ′,𝑠⟩ factors as (𝑇 − 1)(𝑇 + 𝑞𝑠) after
a rescaling of the indeterminate 𝑇 . Let 𝑙 : 𝐼𝑟𝑒𝑓 → Z≥0 denote the length function
of the Coxeter system (𝐼𝑟𝑒𝑓 , 𝑆𝑊 ′,𝐿). We then have the following presentation for the
generalized Hecke algebra ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ′′), analogous to the presentation given in
[46, Theorem 4.14] in the context of finite groups of Lie type:
Theorem 3.3.1.1. Let 𝜇 ∈ 𝑍2(𝐼,C×) be the 2-cocycle appearing in Theorem 3.2.2.4.
There is a basis 𝑇𝑥 : 𝑥 ∈ 𝐼 for ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ′′) with multiplication law completely
described by the following relations for all 𝑥 ∈ 𝐼, 𝑑 ∈ 𝐼𝑐𝑜𝑚𝑝, 𝑤 ∈ 𝐼𝑟𝑒𝑓 , and 𝑠 ∈ 𝑆𝑊 ′,𝐿:
(1) 𝑇𝑑𝑇𝑥 = 𝜇(𝑑, 𝑥)−1𝑇𝑑𝑥
(2) 𝑇𝑥𝑇𝑑 = 𝜇(𝑥, 𝑑)−1𝑇𝑥𝑑
(3) 𝑇𝑠𝑇𝑤 =
⎧⎪⎨⎪⎩𝜇(𝑠, 𝑤)−1𝑇𝑠𝑤 if 𝑙(𝑠𝑤) > 𝑙(𝑤)
𝑞𝑠𝜇(𝑠, 𝑤)−1𝑇𝑠𝑤 + (𝑞𝑠 − 1)𝑇𝑤 if 𝑙(𝑠𝑤) < 𝑙(𝑤)
(4) 𝑇𝑤𝑇𝑠 =
⎧⎪⎨⎪⎩𝜇(𝑤, 𝑠)−1𝑇𝑤𝑠 if 𝑙(𝑤𝑠) > 𝑙(𝑤)
𝑞𝑠𝜇(𝑤, 𝑠)−1𝑇𝑤𝑠 + (𝑞𝑠 − 1)𝑇𝑤 if 𝑙(𝑤𝑠) < 𝑙(𝑤).
Remark 3.3.1.2. When the cocycle 𝜇 is trivial, Theorem 3.3.1.1 gives an isomor-
phism
ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ′′) ∼= 𝐼𝑐𝑜𝑚𝑝 n H𝑞(𝐼𝑟𝑒𝑓 )
where H𝑞(𝐼𝑟𝑒𝑓 ) denotes the Iwahori-Hecke algebra associated with the Coxeter system
(𝐼𝑟𝑒𝑓 , 𝑆𝑊 ′,𝐿) with parameter 𝑞 : 𝑠 ↦→ 𝑞𝑠 and where 𝐼𝑐𝑜𝑚𝑝 acts on H𝑞(𝐼𝑟𝑒𝑓 ) by automor-
phisms induced by diagram automorphisms of the Dynkin diagram of (𝐼𝑟𝑒𝑓 , 𝑆𝑊 ′,𝐿).
We will see in a later section, by inspecting all possible cases, that the cocycle 𝜇 is
indeed trivial in every case.
Proof. Take the base point 𝑏 ∈ h𝑊′
𝑟𝑒𝑔 for the fundamental group to lie in the fundamen-
68
tal chamber 𝒞 ⊂ h𝑊′
R,𝐿−𝑟𝑒𝑔 and outside of the hyperplanes h𝑊′′ ⊂ h𝑊
′ for the 𝐿-trivial
parabolic subgroups 𝑊 ′′ ⊃ 𝑊 ′. As 𝐼 acts freely on h𝑊′
𝑟𝑒𝑔, so too 𝐼𝑐𝑜𝑚𝑝 acts freely on
the orbit 𝐼𝑐𝑜𝑚𝑝.𝑏 ⊂ 𝒞 ∩ h𝑊′
𝑟𝑒𝑔. For each 𝑑 ∈ 𝐼𝑐𝑜𝑚𝑝, choose a path 𝛾𝑑 : [0, 1] → 𝒞 ∩ h𝑊′
𝑟𝑒𝑔
in 𝒞 ∩ h𝑊′
𝑟𝑒𝑔 from 𝑏 to 𝑑.𝑏. Let 𝛾𝑑 denote the reverse path, so that the concatenation
𝛾𝑑 * 𝛾𝑑 is a loop in 𝒞 ∩ h𝑊′
𝑟𝑒𝑔 with base point 𝑏. As 𝒞 is contractible, the image of the
path homotopy class of 𝛾𝑑 * 𝛾𝑑 under the homomorphism
𝜋1(h𝑊 ′
𝑟𝑒𝑔) → 𝜋1(h𝑊 ′
𝐿−𝑟𝑒𝑔)
induced by the inclusion h𝑊′
𝑟𝑒𝑔 ⊂ h𝑊′
𝐿−𝑟𝑒𝑔 is trivial. As discussed in the proof of Theorem
3.2.3.6, the kernel of this map is generated (as a normal subgroup) by the generators
of monodromy 𝑇𝑊 ′′ about the hyperplanes h𝑊 ′′ for the 𝐿-trivial parabolic subgroups
𝑊 ′′ ⊃ 𝑊 ′. It follows immediately that the image of the homotopy class [𝛾𝑑] in the
quotient𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)
⟨𝑇𝑊 ′′ : 𝑊 ′′ is 𝐿-trivial⟩
is independent of the choice of the path 𝛾𝑑 and that the assignments 𝑑 ↦→ [𝛾𝑑] for
𝑑 ∈ 𝐼𝑐𝑜𝑚𝑝 extends uniquely to an algebra homomorphism
C−𝜇[𝐼𝑐𝑜𝑚𝑝] →C−[𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑇𝑊 ′′ : 𝑊 ′′ is 𝐿-trivial⟩.
Composing with the automorphism 𝜏𝜒 of C−[𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼)] discussed in the proof of
Theorem 3.2.3.6, this yields an algebra homomorphism
C−𝜇[𝐼𝑐𝑜𝑚𝑝] →C−[𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩.
We also have the embedding
C[𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′′ is 𝐿-essential⟩→
C−[𝜋1(h𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩
69
from Theorem 3.2.3.6. It is clear that the map of C-vector spaces
C−𝜇[𝐼𝑐𝑜𝑚𝑝] ⊗CC−[𝜋1(h
𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′′ is 𝐿-essential⟩
→C−[𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩
induced by multiplication is surjective. It follows by Theorem 3.2.3.6 that this map
is in fact an isomorphism of C-vector spaces.
For a simple reflection 𝑠 ∈ 𝑆𝑊 ′,𝐿 ⊂ 𝐼𝑟𝑒𝑓 , let 𝑒𝑠 denote the generator of monodromy
𝑇𝑊 ′,𝐿,⟨𝑊 ′,𝑠⟩. For any 𝑤 ∈ 𝐼𝑟𝑒𝑓 , let 𝑒𝑤 ∈ 𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 ) denote the element obtained as
the product 𝑒𝑤 := 𝑒𝑠1 · · · 𝑒𝑠𝑙 for any reduced expression 𝑤 = 𝑠1 · · · 𝑠𝑙 of 𝑤 as a product
of simple reflections in 𝑆𝑊 ′,𝐿, as in the last paragraph of the proof of Theorem 3.2.3.6
(recall that the product here is taken in the fundamental group 𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 ), i.e.
without regards to the cocycle −). For 𝑤 ∈ 𝐼𝑟𝑒𝑓 , let 𝑇𝑤 denote 𝑒𝑤 viewed as an
element of C−[𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]. By Theorem 3.2.3.6 and the last paragraph in its
proof, the set 𝑇𝑤 : 𝑤 ∈ 𝐼𝑟𝑒𝑓 forms a basis for the quotient
C−[𝜋1(h𝑊 ′𝐿−𝑟𝑒𝑔/𝐼
𝑟𝑒𝑓 )]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′′ is 𝐿-essential⟩.
By rescaling the elements 𝑇𝑤 appropriately using twists by characters as 𝜏𝜒 was used
in the proof of Theorem 3.2.3.6, we may assume that the quadratic relations the 𝑇𝑠
satisfy are of the form
(𝑇𝑠 − 1)(𝑇𝑠 + 𝑞𝑠) = 0
for some uniquely determined 𝑞𝑠 ∈ C×. The assignments 𝑠 ↦→ 𝑞𝑠 determine an 𝐼-
invariant function 𝑞 : 𝑆𝑊 ′,𝐿 → C× with 𝑞(𝑠) = 𝑞𝑠. The computations at the end
of the proof of Theorem 3.2.3.6 show that for 𝑠 ∈ 𝑆𝑊 ′,𝐿 and 𝑤 ∈ 𝐼𝑟𝑒𝑓 we have the
multiplication law
𝑇𝑠𝑇𝑤 =
⎧⎪⎨⎪⎩𝜇(𝑠, 𝑤)−1𝑇𝑠𝑤 if 𝑙(𝑠𝑤) > 𝑙(𝑤)
𝑞𝑠𝜇(𝑠, 𝑤)−1𝑇𝑠𝑤 + (𝑞𝑠 − 1)𝑇𝑤 if 𝑙(𝑠𝑤) < 𝑙(𝑤)
70
where 𝑙 : 𝐼𝑟𝑒𝑓 → Z≥0 is the length function determined by the choice of simple reflec-
tions 𝑆𝑊 ′,𝐿 (note that 𝜇(𝑠, 𝑠𝑤) = 𝜇(𝑠, 𝑤)−1 because 𝑠2 = 1). An entirely analogous
calculation shows that
𝑇𝑤𝑇𝑠 =
⎧⎪⎨⎪⎩𝜇(𝑤, 𝑠)−1𝑇𝑤𝑠 if 𝑙(𝑤𝑠) > 𝑙(𝑤)
𝑞𝑠𝜇(𝑤, 𝑠)−1𝑇𝑤𝑠 + (𝑞𝑠 − 1)𝑇𝑤 if 𝑙(𝑤𝑠) < 𝑙(𝑤).
For 𝑑 ∈ 𝐼𝑐𝑜𝑚𝑝, let
𝑒𝑑 ∈𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)
⟨𝑇𝑊 ′′ : 𝑊 ′′ is 𝐿-trivial⟩
be the image of [𝛾𝑑] as constructed above. We may also regard the 𝑒𝑤 for 𝑤 ∈ 𝐼𝑟𝑒𝑓 as
elements of this quotient. Any element 𝑥 ∈ 𝐼 can be written uniquely as a product
𝑥 = 𝑑𝑤 for some 𝑑 ∈ 𝐼𝑐𝑜𝑚𝑝 and 𝑤 ∈ 𝐼𝑟𝑒𝑓 , and we may therefore define 𝑒𝑥 := 𝑒𝑑𝑒𝑤.
Note that 𝑒𝑑𝑒𝑤𝑒𝑑−1 = 𝑒𝑑𝑤𝑑−1 so 𝑒𝑑𝑤𝑑−1𝑒𝑑 = 𝑒𝑥, so writing 𝑥 = 𝑤′𝑑′ with 𝑤′ ∈ 𝐼𝑟𝑒𝑓
and 𝑑′ ∈ 𝐼𝑐𝑜𝑚𝑝 and taking the element 𝑒𝑤′𝑒𝑑′ defines the same element 𝑒𝑥. Note that
𝑒𝑑𝑒′𝑑 = 𝑒𝑑𝑑′ for any 𝑑, 𝑑′ ∈ 𝐼𝑐𝑜𝑚𝑝, and hence 𝑒𝑑𝑒𝑥 = 𝑒𝑑𝑥 and 𝑒𝑥𝑒𝑑 = 𝑒𝑥𝑑 for any 𝑑 ∈ 𝐼𝑐𝑜𝑚𝑝
and 𝑤 ∈ 𝐼. For any 𝑥 ∈ 𝐼, let 𝑇𝑥 denote the element 𝑒𝑥 viewed as an element of the
quotient algebraC−[𝜋1(h
𝑊 ′𝑟𝑒𝑔/𝐼)]
⟨𝑃𝑊 ′′(𝑇𝑊 ′′) : 𝑊 ′ is corank 1 in 𝑊 ′′⟩.
For 𝑥 ∈ 𝐼𝑟𝑒𝑓 , the elements 𝑇𝑥 are the images of the elements 𝑇𝑥 considered above.
The multiplication rules 𝑇𝑑𝑇𝑥 = 𝜇(𝑑, 𝑥)−1𝑇𝑑𝑥 and 𝑇𝑥𝑇𝑑 = 𝜇(𝑥, 𝑑)−1𝑇𝑥𝑑 for 𝑑 ∈ 𝐼𝑐𝑜𝑚𝑝
and 𝑥 ∈ 𝐼 follow immediately from the computations with 𝑒𝑥 and 𝑒𝑑 above and the
definition of the twisted group algebra. That 𝑇𝑥 : 𝑥 ∈ 𝐼 forms a basis for this
quotient, and hence also for ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ′′) by Theorem 3.2.3.6 follows immediately
from the considerations above, and the theorem follows.
3.3.2 Computing Parameters Via KZ
To compute the quadratic relations that the generators 𝑇𝑠 ∈ ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) satisfy,
we will reduce the problem to certain explicit computations in the Hecke algebra
H𝑞(𝑊 ) of the ambient group 𝑊 and its representations. This is possible thanks to
71
the following result of Shan:
Lemma 3.3.2.1. [65, Lemma 2.4] Let 𝐾𝑍 : 𝒪𝑐(𝑊, h) → H𝑞(𝑊 )-mod𝑓.𝑑. be the 𝐾𝑍
functor, and let 𝐾,𝐿 : 𝒪𝑐(𝑊, h) → 𝒪𝑐(𝑊, h) be two right exact functors that map
projective objects to projective objects. Then the natural map of vector spaces
Hom(𝐾,𝐿) → Hom(𝐾𝑍 ∘𝐾,𝐾𝑍 ∘ 𝐿)
𝑓 ↦→ 1𝐾𝑍𝑓
is an isomorphism.
Assume that (𝑊 ′, 𝑆 ′) ⊂ (𝑊,𝑆) is a Coxeter subsystem of corank 1. The following
lemma describing the canonical complement 𝑁𝑊 ′ to 𝑊 ′ in 𝑁𝑊 (𝑊 ′) will be useful in
what follows:
Lemma 3.3.2.2. Either 𝑊 ′ is self-normalizing in 𝑊 or has index 2 in its normalizer
𝑁𝑊 (𝑊 ′). In the latter case, the longest elements 𝑤0 and 𝑤′0 of 𝑊 and 𝑊 ′, respectively,
commute, and the canonical complement 𝑁𝑊 ′ to 𝑊 ′ in 𝑁𝑊 (𝑊 ′) is
𝑁𝑊 ′ = 1, 𝑤0𝑤′0.
Proof. The first statement follows from Lemma 3.2.3.3. Let 𝑤 ∈ 𝑁𝑊 ′ be the nontrivial
element. Let Φ ⊂ hR denote the root system of𝑊 , let Φ𝑊 ′ ⊂ Φ denote the root system
of 𝑊 ′, let 𝛼1, ..., 𝛼𝑟 be an ordering of the simple roots in Φ so that 𝛼1, ..., 𝛼𝑟−1 are the
simple roots in Φ𝑊 ′ , and let Φ+ ⊂ Φ and Φ+𝑊 ′ ⊂ Φ𝑊 ′ denote the respective subsets
of positive roots. By definition of 𝑁𝑊 ′ , 𝑤(Φ+𝑊 ′) = Φ+
𝑊 ′ . From the proof of Lemma
3.2.3.3, we see that 𝑤 acts by −1 in h𝑊′
R . As the remaining simple root 𝛼𝑟 is the
unique simple root with nonzero component in h𝑊′
R with respect to the decomposition
hR = hR,𝑊 ′⊕h𝑊′
R and as every positive root 𝛼 ∈ Φ+ is uniquely of the form 𝛼 =∑
𝑖 𝑛𝑖𝛼𝑖
for some nonnegative integers 𝑛𝑖 ≥ 0, it follows that 𝑤(Φ+∖Φ+𝑊 ′) ⊂ Φ−. It follows
that the inversion set of 𝑤 is precisely Φ+∖Φ+𝑊 ′ and hence that 𝑤 = 𝑤0𝑤
′0. As 𝑤,𝑤0,
and 𝑤′0 are involutions, it follows that 𝑤0 and 𝑤′
0 commute.
72
As in previous sections, let 𝑐 : 𝑆 → C be a 𝑊 -invariant function and let 𝐿 be an
irreducible finite-dimensional representation of 𝐻𝑐(𝑊′, h𝑊 ′). Let 𝑤0 and 𝑤′
0 denote
the longest elements in 𝑊 and 𝑊 ′, respectively. We will assume that 𝑊 ′ is not self-
normalizing in 𝑊 , and by Lemma 3.3.2.2 it follows that 𝑤0 and 𝑤′0 commute, that 𝑊 ′
is index 2 in 𝑁𝑊 ′(𝑊 ), and that 𝑁𝑊 ′ = 1, 𝑤0𝑤′0. We will assume that the involution
𝑤0𝑤′0 fixes the isomorphism class of 𝐿 so that 𝐼 = 𝑁𝑊 ′ and the monodromy operator
𝑇𝑊 ′ satisfies a nontrivial quadratic relation as in Theorem 3.2.3.6.
The following observation in this setting will be central to our approach as it will
allow for the explicit computation, via computations in the Hecke algebra H𝑞(𝑊 ), of
the eigenvalues of monodromy in the local systems arising from the functors 𝐾𝑍𝐿.
This lemma should be regarded as a generalization to Coxeter groups of arbitrary
type of the calculation appearing in [40, Lemma 4.14].
Lemma 3.3.2.3. Let 𝒞𝑊 ′ ⊂ h𝑊 ′,R be the open fundamental Weyl chamber associated
to (𝑊 ′, 𝑆 ′), let 𝒞𝑊 ⊂ hR be the open fundamental Weyl chamber associated to (𝑊,𝑆),
and choose a basepoint 𝑏 = (𝑏′, 𝑏′′) ∈ 𝒞𝑊 ′ × h𝑊′
𝑟𝑒𝑔 lying in 𝒞𝑊 . As 𝒞𝑊 ′ is contractible
and stable under 𝑤0𝑤′0, the pair (𝛾, 𝑇𝑊 ′), where 𝛾 is any path in 𝒞𝑊 ′ from 𝑏′ to
𝑤0𝑤′0𝑏
′ and 𝑇𝑊 ′ is the half-loop in h𝑊′
𝑟𝑒𝑔 lifting the canonical generator of monodromy
in 𝜋1(h𝑊 ′𝑟𝑒𝑔/𝑁𝑊 ′), determines an element in 𝜋1(((h𝑊 ′)𝑟𝑒𝑔 × h𝑊
′𝑟𝑒𝑔)/𝑁𝑊 ′) that does not
depend on the choice of 𝛾. The image of this element under the natural map
𝜄𝑊 ′,1 : 𝜋1(((h𝑊 ′)𝑟𝑒𝑔 × h𝑊′
𝑟𝑒𝑔)/𝑁𝑊 ′) → 𝜋1(h𝑟𝑒𝑔/𝑊 ) = 𝐵𝑊
is 𝑇𝑤0𝑤′0.
Proof. That the element of 𝜋1(((h𝑊 ′)𝑟𝑒𝑔×h𝑊′
𝑟𝑒𝑔)/𝑁𝑊 ′) determined by the pair (𝛾, 𝑇𝑊 ′)
does not depend on the choice of 𝛾 follows immediately from the contractibility of
the fundamental Weyl chamber 𝒞𝑊 ′ . By definition of 𝜄𝑊 ′,1, for any sufficiently large
real number 𝑅 > 0 the image 𝜄𝑊 ′,1(𝛾, 𝑇𝑊 ′) in 𝜋1(h𝑟𝑒𝑔/𝑊 ) is represented by the image
of the path
𝑝 : [0, 1] → h𝑟𝑒𝑔
73
𝑝(𝑡) = (𝛾(𝑡), 𝑅𝑒𝜋𝑖𝑡𝑏′′)
under the natural projection h𝑟𝑒𝑔 → h𝑟𝑒𝑔/𝑊 . As 𝑤0𝑤′0𝑏
′′ = −𝑏′′, it follows that 𝑝 is a
path from the point 𝑝(0) = (𝑏′, 𝑅𝑏′′) ∈ 𝒞𝑊 to the point 𝑝(1) = 𝑤0𝑤′0𝑝(0) ∈ 𝑤0𝑤
′0𝒞𝑊 .
For each positive root 𝛼 ∈ Φ+, let 𝐻𝛼 := ker(𝛼) ⊂ h be the associated reflection
hyperplane. That 𝑝 represents the element 𝑇𝑤0𝑤′0
follows from the observation that
𝑝 traverses a positively-oriented (with respect to the complex structure) half-loop
about each hyperplane 𝐻𝛼 for roots 𝛼 ∈ Φ+∖Φ+𝑊 ′ while 𝑝 does not encircle any of the
remaining hyperplanes 𝐻𝛼 for roots 𝛼 ∈ Φ+𝑊 ′ . More precisely, for roots 𝛼 ∈ Φ+∖Φ+
𝑊 ′
the composition 𝛼∘𝑝 : [0, 1] → C× determines a path from the positive real axis R+ to
the negative real axis R− lying entirely in the upper half-space 𝑧 ∈ C× : Re(𝑧) ≥ 0.
For roots 𝛼 ∈ Φ+𝑊 ′ , the composition 𝛼 ∘ 𝑝 : [0, 1] → C× determines a path lying
entirely on R+, as 𝛾(𝑡) ∈ 𝒞𝑊 ′ and 𝛼(𝑏′′) = 0. The equality 𝜄𝑊 ′,1(𝛾, 𝑇𝑊 ′) = 𝑇𝑤0𝑤′0
follows.
Let Mon(𝑇𝑊 ) denote the isomorphism of functors
Res𝑊𝑊 ′ → tw𝑤0𝑤′0∘ Res𝑊𝑊 ′
arising from monodromy along the generator of monodromy 𝑇𝑊 in the local system
Res𝑊𝑊 ′ , where Res𝑊𝑊 ′ is the partial 𝐾𝑍 functor recalled in Section 2.5. The following
is an immediate corollary of Lemma 3.3.2.3 and the transitivity result of Gordon-
Martino recalled in Theorem 2.6.0.1:
Lemma 3.3.2.4. Multiplication by 𝑇𝑤0𝑤′0
defines an isomorphism
𝑇𝑤0𝑤′0
: 𝐻Res𝑊𝑊 ′ → 𝐻tw𝑤0𝑤′0∘ 𝐻Res𝑊𝑊 ′
of functors H𝑞(𝑊′)-mod𝑓.𝑑. → H𝑞(𝑊
′)-mod𝑓.𝑑.. Furthermore, Mon(𝑇𝑊 ) is the lift to
𝒪𝑐(𝑊, h) of 𝑇𝑤0𝑤′0
in the sense that, with respect to the identifications 𝐾𝑍 ∘Res𝑊𝑊 ′ =
𝐻Res𝑊𝑊 ′ ∘𝐾𝑍 and 𝐾𝑍 ∘ tw𝑤0𝑤′0
= 𝐻tw𝑤0𝑤′0∘𝐾𝑍, we have an equality
1𝐾𝑍Mon(𝑇𝑊 ) = 𝑇𝑤0𝑤′01𝐾𝑍
74
of isomorphisms of functors
𝐾𝑍 ∘ Res𝑊𝑊 ′ → 𝐾𝑍 ∘ tw𝑤0𝑤′0∘ Res𝑊𝑊 ′ .
Remark 3.3.2.5. When the meaning is clear, we denote the 𝐾𝑍 functors defined for
𝒪𝑐(𝑊, h) and 𝒪𝑐(𝑊′, h𝑊 ′) each by 𝐾𝑍.
Proof of Lemma 3.3.2.4. The first statement follows from the observation that the
functor 𝐻Res𝑊𝑊 ′ is represented by the H𝑞(𝑊′)-H𝑞(𝑊 )-bimodule H𝑞(𝑊 ) and the ob-
servation that 𝑇𝑤0𝑤′0𝑇𝑠 = 𝑇𝜎𝑤0𝑤
′0(𝑠)𝑇𝑤0𝑤′
0for all 𝑠 ∈ 𝑆 ′, where 𝜎𝑤0𝑤′
0is the diagram
automorphism of (𝑊 ′, 𝑆 ′) induced by conjugation by 𝑤0𝑤′0 in 𝑊 . The second state-
ment is an immediate consequence of Lemma 3.3.2.3 and Theorem 2.6.0.1.
Definition 3.3.2.6. Let 𝑋𝑊 ′ ⊂ 𝑊 be the set of minimal length left 𝑊 ′-coset repre-
sentatives, and let 𝑧𝑑𝑑∈𝑋𝑊 ′ be the unique set of elements 𝑧𝑑 ∈ H𝑞(𝑊′) such that
𝑇 2𝑤0𝑤′
0=∑𝑑∈𝑋𝑊 ′
𝑧𝑑𝑇𝑑.
We are interested in the elements 𝑧1, 𝑧𝑤0𝑤′0∈ H𝑞(𝑊
′) arising from the left 𝑊 ′-
cosets that are also right 𝑊 ′-cosets. We now establish important properties that
these elements satisfy:
Lemma 3.3.2.7. The element 𝑧𝑤0𝑤′0
commutes with 𝑇𝑤0𝑤′0
and satisfies the relation
𝑥𝑧𝑤0𝑤′0
= 𝑧𝑤0𝑤′0𝜎(𝑥)
for all 𝑥 ∈ H𝑞(𝑊′), where 𝜎 : H𝑞(𝑊
′) → H𝑞(𝑊′) is the automorphism of H𝑞(𝑊
′)
arising from the diagram automorphism of (𝑊 ′, 𝑆 ′) obtained by conjugation by 𝑤0𝑤′0
in 𝑊 . In particular, multiplication by 𝑧𝑤0𝑤′0
defines a morphism
𝑧𝑤0𝑤′0
: Id → 𝐻tw𝑤0𝑤′0
75
of functors H𝑞(𝑊′)-mod𝑓.𝑑. → H𝑞(𝑊
′)-mod𝑓.𝑑.. Furthermore,
𝑧1 = 𝑞𝑙(𝑤0𝑤′0),
where, in the unequal parameter case, 𝑞𝑙(𝑤0𝑤′0) denotes the product
𝑞𝑙(𝑤0𝑤′0) :=
∏𝑠∈𝑆/𝑊
𝑞𝑙𝑠(𝑤0𝑤′0)
𝑠
over the conjugacy classes of reflections in 𝑊 , where 𝑙𝑠 : 𝑊 → Z≥0 is the length
function of 𝑊 attached to the conjugacy class of 𝑠 and 𝑞𝑠 ∈ C is the parameter
associated to the conjugacy class of 𝑠.
Proof. As 𝑤0 and 𝑤′0 commute, we have 𝑇𝑤0𝑇
−1𝑤′
0= 𝑇𝑤0𝑤′
0= 𝑇𝑤′
0𝑤0= 𝑇−1
𝑤′0𝑇𝑤0 . Every
simple reflection in 𝑆 lies in both the right and left descent sets of 𝑤0, and similarly
every simple reflection in 𝑆 ′ lies in both the right and left descent sets of 𝑤′0. It follows
that 𝑇𝑤0𝑇𝑠 = 𝑇𝜎𝑤0 (𝑠)𝑇𝑤0 for every 𝑠 ∈ 𝑆, where 𝜎𝑤0 is the diagram automorphism of
(𝑊,𝑆) obtained by conjugation by 𝑤0, and similarly 𝑇𝑤′0𝑇𝑠 = 𝑇𝜎𝑤′
0(𝑠)𝑇𝑤′
0for all 𝑠 ∈ 𝑆 ′,
where 𝜎𝑤′0is the diagram automorphism of (𝑊 ′, 𝑆 ′) obtained by conjugation by 𝑤′
0. It
follows that 𝑇 2𝑤0
is central in H𝑞(𝑊 ) and 𝑇 2𝑤′
0is central in H𝑞(𝑊
′) (this is well known
[6, 17]) and that 𝑇𝑤0𝑤′0𝑥 = 𝜎(𝑥)𝑇𝑤0𝑤′
0for all 𝑥 ∈ H𝑞(𝑊
′). In particular, the element
𝑇 2𝑤0𝑤′
0= 𝑇 2
𝑤0𝑇−2𝑤′
0∈ H𝑞(𝑊 ) considered above centralizes H𝑞(𝑊
′). As multiplication
by elements of H𝑞(𝑊′) on the right or left respects the decomposition of 𝑇 2
𝑤0𝑤′0
by
𝑊 ′-double cosets and as 𝑇𝑤0𝑤′0
normalizes H𝑞(𝑊′), it follows that 𝑥𝑧𝑤0𝑤′
0𝑇𝑤0𝑤′
0=
𝑧𝑤0𝑤′0𝑇𝑤0𝑤′
0𝑥 for all 𝑥 ∈ H𝑞(𝑊
′). But as 𝑇𝑤0𝑤′0𝑥 = 𝜎(𝑥)𝑇𝑤0𝑤′
0and 𝑇𝑤0𝑤′
0is invertible,
it follows that 𝑥𝑧𝑤0𝑤′0
= 𝑧𝑤0𝑤′0𝜎(𝑥). That multiplication by 𝑧𝑤0𝑤′
0defines a morphism
of functors Id → 𝐻tw𝑤0𝑤′0
follows immediately.
Similarly, note that conjugation by 𝑇𝑤′0
clearly respects decomposition of ele-
ments by 𝑊 ′-double cosets and that conjugation by 𝑇𝑤0 stabilizes H𝑞(𝑊′) and sends
minimal length 𝑊 ′-double-coset representatives of to minimal length 𝑊 ′-double-
coset representatives of the same length. In particular, conjugation by 𝑇𝑤0𝑤′0
fixes
the top degree term 𝑧𝑤0𝑤′0𝑇𝑤0𝑤′
0in the decomposition of 𝑇 2
𝑤0𝑤′0
by left 𝑊 ′-cosets.
76
As 𝑇𝑤0𝑤′0
commutes with itself and 𝑇𝑤0𝑤′0𝑧𝑤0𝑤′
0= 𝜎(𝑧𝑤0𝑤′
0)𝑇𝑤0𝑤′
0this implies that
𝑧𝑤0𝑤′0𝑇𝑤0𝑤′
0= 𝜎(𝑧𝑤0𝑤′
0)𝑇𝑤0𝑤′
0and hence that 𝜎(𝑧𝑤0𝑤′
0) = 𝑧𝑤0𝑤′
0. In particular, 𝑧𝑤0𝑤′
0
commutes with 𝑇𝑤0𝑤′0.
Finally, to show that 𝑧1 = 𝑞𝑙(𝑤0𝑤′0), by multiplying on the left by 𝑇𝑤′
0it suffices to
show that the component of 𝑇𝑤0𝑇𝑤0𝑤′0lying in H𝑞(𝑊
′) according to the decomposition
of H𝑞(𝑊 ) by left 𝑊 ′-cosets is 𝑞𝑙(𝑤0𝑤′0)𝑇𝑤′
0. This is clear from the interaction of the
multiplication laws defining H𝑞(𝑊 ) and the length function.
Definition 3.3.2.8. Let 𝐶𝑧𝑤0𝑤′0
be the unique morphism
𝐶𝑧𝑤0𝑤′0
: Id → tw𝑤0𝑤′0
of functors 𝒪𝑐(𝑊′, h𝑊 ′) → 𝒪𝑐(𝑊
′, h𝑊 ′) lifting 𝑧𝑤0𝑤′0
in the sense of Lemma 3.3.2.1,
i.e. so that
1𝐾𝑍𝐶𝑧𝑤0𝑤′
0= 𝑧𝑤0𝑤′
01𝐾𝑍
with respect to the identification 𝐾𝑍 ∘ 𝑡𝑤𝑤0𝑤′0
= 𝐻tw𝑤0𝑤′0∘𝐾𝑍.
Definition 3.3.2.9. Let 𝜇𝑤0𝑤′0
be the isomorphism of functors
𝜇𝑤0𝑤′0
: Id ⊕ tw𝑤0𝑤′0→ tw𝑤0𝑤′
0∘ (Id ⊕ tw𝑤0𝑤′
0) = tw𝑤0𝑤′
0⊕ Id
defined by the matrix
𝜇𝑤0𝑤′0
:=
⎛⎝0 𝑞𝑙(𝑤0𝑤′0)
1 𝐶𝑧𝑤0𝑤′0
⎞⎠ .
Recall that by the Mackey formula for rational Cherednik algebras attached to
Coxeter groups (see Section 2.7), and the fact that 𝐿 is finite-dimensional and hence
annihilated by all restriction functors Res𝑊′
𝑊 ′′ for proper parabolic subgroups 𝑊 ′′ (
𝑊 ′, that we have
(Res𝑊𝑊 ′ ∘ Ind𝑊𝑊 ′)𝐿 ∼= (Id ⊕ tw𝑤0𝑤′0)𝐿.
Lemma 3.3.2.10. With respect to the isomorphism
(Res𝑊𝑊 ′ ∘ Ind𝑊𝑊 ′)𝐿 ∼= (Id ⊕ tw𝑤0𝑤′0)𝐿
77
arising from the Mackey formula, the isomorphisms
(Mon(𝑇𝑊 )1Ind𝑊𝑊 ′
)(𝐿), 𝜇(𝐿) : (Id ⊕ tw𝑤0𝑤′0)𝐿→ (tw𝑤0𝑤′
0⊕ Id)𝐿
are equal.
Proof. This is an immediate consequence of the compatibility of the Mackey decom-
position for rational Cherednik algebras with the 𝐾𝑍 functor, Definition 3.3.2.6, and
Lemmas 3.3.2.4 and 3.3.2.7.
Notation: For a central element 𝑧 ∈ H𝑞(𝑊′), let 𝑧|𝐿 ∈ C denote the scalar by which
𝑧 acts on irreducible representations of H𝑞(𝑊 ′) lying in the block of H𝑞(𝑊 ′)-mod𝑓.𝑑.
corresponding to the block of 𝒪𝑐(𝑊′, h𝑊 ′) containing 𝐿 under the 𝐾𝑍 functor.
We can now give a formula for the quadratic relations satisfied by the elements
𝑇𝑠 appearing in Theorem 3.3.1.1:
Theorem 3.3.2.11. Let 𝑇 denote the element 𝑇𝑊 ∈ ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ), and let 𝑛 ∈
AutC(𝐿) denote the involution of 𝐿 by which 𝑤0𝑤′0 ∈ 𝑁𝑊 ′ acts making 𝐿 𝑁𝑊 ′-
equivariant as a 𝐻𝑐(𝑊′, h𝑊 ′)-module. Then 𝑇 satisfies the quadratic relation
𝑇 2 = (𝐶𝑧𝑤0𝑤′0(𝐿) ∘ 𝑛)|𝐿𝑇 + 𝑞𝑙(𝑤0𝑤′
0)
where (𝐶𝑧𝑤0𝑤′0(𝐿) ∘ 𝑛)|𝐿 ∈ C denotes the scalar by which the 𝐻𝑐(𝑊
′, h𝑊 ′)-module
homomorphism 𝑛 ∘ 𝐶𝑧𝑤0𝑤′0(𝐿) acts on the irreducible representation 𝐿.
Furthremore, if the diagram automorphism 𝜎 = 𝜎𝑤0𝑤′0
of (𝑊 ′, 𝑆 ′) arising from
conjugation by 𝑤0𝑤′0 is trivial, 𝑇 satisfies the quadratic relation
𝑇 2 = (𝑧𝑤0𝑤′0)|𝐿𝑇 + 𝑞𝑙(𝑤0𝑤′
0).
If the diagram automorphism 𝜎𝑤0 is trivial but the diagram automorphism 𝜎𝑤′0
is
nontrivial, 𝑇 satisfies the quadratic relation
𝑇 2 = (𝑧𝑤0𝑤′0𝑇𝑤′
0)|𝐿(𝑇 2
𝑤′0)|−1/2𝐿 𝑇 + 𝑞𝑙(𝑤0𝑤′
0).
78
Remark 3.3.2.12. The projective representation of 𝑁𝑊 ′ = 𝐼 ∼= Z/2Z on 𝐿 lifts to
an linear representation of 𝑁𝑊 ′ in two ways, differing by a tensor product with the
nontrivial character of 𝑁𝑊 ′, so there is a choice of sign for the action of the nontrivial
element 𝑤0𝑤′0 ∈ 𝑁𝑊 ′ on 𝐿. The quadratic relations appearing in Theorem 3.3.2.11
under assumptions on the diagram automorphisms hold for an appropriate choice of
sign for the operator 𝑛 - choosing the other sign simply negates the linear term in
the quadratic relation. The quadratic relations of the form (𝑇𝑠 − 1)(𝑇𝑠 + 𝑞𝑠) = 0
appearing in Theorem 3.3.1.1 are obtained by rescaling the generators 𝑇𝑊 by twisting
by characters of 𝜋1(h𝑊′
𝑟𝑒𝑔/𝐼) as in Lemma 3.2.3.8, and the relations in this normalized
form are only determined up to inverting 𝑞𝑠 but do not depend on the choice of sign
for the action of 𝑤0𝑤′0 on 𝐿.
Remark 3.3.2.13. The case in which the diagram automorphism 𝜎𝑤0 is nontrivial
but the diagram automorphism 𝜎𝑤′0
is trivial only appears for groups of type 𝐷. We
will show later in Section 3.3.5 how to reduce the problem of computing the quadratic
relations in type 𝐷 to the type 𝐵 case in which this complication does not arise.
Proof. By Proposition 3.2.1.4, Ind𝑊𝑊 ′𝐿 is a projective generator of 𝒪𝑐,𝑊 ′,𝐿, and hence
it follows from Theorem 3.2.2.4 that the action of ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) on the hom space
Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,Res𝑊𝑊 ′Ind𝑊𝑊 ′𝐿) is faithful. It therefore suffices to check that the
quadratic relation holds in this representation.
It follows from Lemma 3.3.2.10 that the action of 𝑇 in the representation
Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,Res𝑊𝑊 ′Ind𝑊𝑊 ′𝐿) ⊂ 𝐿* ⊗C Res𝑊𝑊 ′Ind𝑊𝑊 ′𝐿
is by the operator
𝑛* ⊗ 𝜇𝑤0𝑤′0(𝐿) = 𝑛* ⊗
⎛⎝0 𝑞𝑙(𝑤0𝑤′0)
1 𝐶𝑧𝑤0𝑤′0(𝐿)
⎞⎠ .
79
As 𝑛2 = 1, a simple calculation yields
𝑇 2 =
⎛⎝𝑛* ⊗
⎛⎝𝐶𝑧𝑤0𝑤′0(𝐿) 0
0 𝐶𝑧𝑤0𝑤′0(𝐿)
⎞⎠⎞⎠ ∘ 𝑇 +
⎛⎝𝑞𝑙(𝑤0𝑤′0) 0
0 𝑞𝑙(𝑤0𝑤′0)
⎞⎠ .
As the operator appearing in front of 𝑇 on the righthand side acts on the hom space
Hom𝒪𝑐(𝑊 ′,h𝑊 ′ )(𝐿,𝐿⊕ tw𝑤0𝑤′0𝐿) by the scalar (𝐶𝑧𝑤0𝑤′
0(𝐿)∘𝑛)|𝐿, the first claim follows.
Now, suppose the diagram automorphism 𝜎𝑤0𝑤′0
of (𝑊 ′, 𝑆 ′) is trivial, so that
𝑤0𝑤′0 centralizes 𝑊 ′ and acts on h𝑊 ′ trivially. In particular, 𝑤0𝑤
′0 acts trivially on
𝐻𝑐(𝑊′, h𝑊 ′), and hence the trivial action of 𝑁𝑊 ′ on 𝐿 makes 𝐿 𝑁𝑊 ′-equivariant, so
we may take 𝑛 = Id𝐿. The quadratic relation for 𝑇 in this case follows immediately.
Finally, suppose the diagram automorphism 𝜎𝑤0 is trivial but the diagram auto-
morphism 𝜎𝑤′0
is not. It follows that the diagram automorphisms 𝜎𝑤′0
and 𝜎 = 𝜎𝑤0𝑤′0
are equal and that 𝑇𝑤′0𝑥 = 𝜎(𝑥)𝑇𝑤′
0for all 𝑥 ∈ H𝑞(𝑊
′). In particular, multiplication
by 𝑇𝑤′0
defines a morphism
𝑇𝑤′0
: Id → 𝐻tw𝑤0𝑤′0
of functors H𝑞(𝑊′)-mod𝑓.𝑑. → H𝑞(𝑊
′)-mod𝑓.𝑑.. Let 𝐶𝑇𝑤′0
be the morphism
𝐶𝑇𝑤′0
: Id → tw𝑤0𝑤′0
of functors 𝒪𝑐(𝑊′, h𝑊 ′) → 𝒪𝑐(𝑊
′, h𝑊 ′) obtained by lifting 𝑇𝑤′0
by Lemma 3.3.2.1,
similarly to the definition of 𝐶𝑧𝑤0𝑤′0
(Definition 3.3.2.8). We may then take the oper-
ator 𝑛 ∈ AutC(𝐿) by which 𝑤0𝑤′0 acts to be the involutive operator
𝑛 = (𝑇 2𝑤′
0)|−1/2𝐿
𝐶𝑇𝑤′0(𝐿).
We then have
(𝐶𝑧𝑤0𝑤′0(𝐿) ∘ 𝑛)|𝐿 = (𝐶𝑧𝑤0𝑤′
0(𝐿) ∘ (𝑇 2
𝑤′0)|−1/2𝐿 (𝐶𝑇𝑤′
0(𝐿)))|𝐿 = (𝑧𝑤0𝑤′
0𝑇𝑤′
0)|𝐿(𝑇 2
𝑤′0)|−1/2𝐿
and the final claim follows.
80
We will see that, in the setting of Coxeter groups, the projective representation of 𝐼
on 𝐿 always lifts to a linear representation, and in particular the cocycle 𝜇 ∈ 𝑍2(𝐼,C×)
is always trivial. Furthermore, the inertia group 𝐼 is always as large as possible, i.e.
it equals 𝑁𝑊 ′ . These facts make the presentations of the algebras ℋ(𝑐,𝑊 ′, 𝐿,𝑊 )
particularly simple:
Theorem 3.3.2.14. Let 𝑊 be a finite Coxeter group with simple reflections 𝑆, let
𝑐 : 𝑆 → C be a class function, let 𝑊 ′ be a standard parabolic subgroup generated by the
simple reflections 𝑆 ′, and let 𝐿 be an irreducible finite-dimensional representation of
the rational Cherednik algebra 𝐻𝑐(𝑊′, h𝑊 ′). Let 𝑁𝑊 ′ denote the canonical complement
to 𝑊 ′ in its normalizer 𝑁𝑊 (𝑊 ′), let 𝑆𝑊 ′ ⊂ 𝑁𝑊 ′ denote the set of reflections in 𝑁𝑊 ′
with respect to its representation in the fixed space h𝑊′, and let 𝑁 𝑟𝑒𝑓
𝑊 ′ denote the
reflection subgroup of 𝑁𝑊 ′ generated by 𝑆𝑊 ′. Let 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ be a complement for 𝑁 𝑟𝑒𝑓
𝑊 ′ in
𝑁𝑊 ′, given as the stabilizer of a choice of fundamental Weyl chamber for the action
of 𝑁 𝑟𝑒𝑓𝑊 ′ on h𝑊
′. Then there is a class function 𝑞𝑊 ′,𝐿 : 𝑆𝑊 ′ → C× and an isomorphism
of algebras
ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) ∼= 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ n H𝑞𝑊 ′,𝐿
(𝑁 𝑟𝑒𝑓𝑊 ′ )
where the semidirect product is defined by the action of 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ on H𝑞𝑊 ′,𝐿
(𝑁 𝑟𝑒𝑓𝑊 ′ ) by
diagram automorphisms arising from the conjugation action of 𝑁 𝑐𝑜𝑚𝑝𝑊 ′ on 𝑁 𝑟𝑒𝑓
𝑊 ′ .
Theorem 3.3.2.14 is proved by case-by-case analysis of the inertia subgroups 𝐼 and
their 2-cocycles 𝜇, to which the rest of this chapter is dedicated. The class function
𝑞𝑊 ′,𝐿 can be explicitly computed using the corank-1 methods developed in Section
3.3.2. We compute this class function in many cases, leading to complete lists of the
irreducible finite-dimensional representations of the algebras 𝐻𝑐(𝑊, h) in many new
cases in exceptional types.
Remark 3.3.2.15. In all cases that we have computed explicitly, the class function
𝑞𝑊 ′,𝐿 depends only on the parabolic subgroup 𝑊 ′, and not on the finite-dimensional
irreducible representation 𝐿. It would be interesting to have a conceptual explanation
for this fact.
81
Notation For a Coxeter group 𝑊 , corank 1 parabolic subgroup 𝑊 ′ ⊂ 𝑊 , and finite-
dimensional irreducible representation 𝐿 of 𝐻𝑐(𝑊′, h𝑊 ′), let 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ) ∈ C×
denote a scalar such that the element 𝑇𝑊 ′,𝐿,𝑊 ∈ ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ), after an appropriate
rescaling, satisfies the quadratic relation
(𝑇 − 1)(𝑇 + 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 )) = 0.
Note that 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ) is determined only up to taking an inverse. The calculations
in the proof of Theorem 3.3.2.11 show that the constant term of the monic quadratic
relation satisfied by the canonical (up to sign) element 𝑇𝑊 ′,𝐿,𝑊 is 𝑞𝑙(𝑤0𝑤′0), and the
linear term in the monic quadratic relation satisfied by 𝑇𝑊 ′,𝐿,𝑊 can therefore be recov-
ered, again up to sign, from 𝑞𝑙(𝑤0𝑤′0) and 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ). Note also that the ambiguity
of 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ) up to inverse has no impact on the isomorphism class of any Iwahori-
Hecke algebra for which 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ) is a parameter, and an explicit isomorphism
can be obtained by scaling the corresponding generators by −𝑞(𝑐,𝑊 ′, 𝐿,𝑊 )−1.
3.3.3 Type 𝐴
Let 𝑛 ≥ 1 be a positive integer, 𝑆𝑛 be the symmetric group on 𝑛 letters with irreducible
reflection representation h = (𝑧1, ..., 𝑧𝑛) ∈ C𝑛 :∑
𝑖 𝑧𝑖 = 0, 𝑐 ∈ C be a complex
number, and let 𝐻𝑐(𝑆𝑛, h) be the associated rational Cherednik algebra. As a first
illustration of the results obtained in the previous sections, let us now recover the
following result of Wilcox describing the subquotients of the filtration of category
𝒪𝑐(𝑆𝑛, h) by the dimension of supports:
Theorem 3.3.3.1. (Wilcox, [73, Theorem 1.8]) Suppose 𝑐 = 𝑟/𝑒 > 0 is a posi-
tive rational number with 𝑟 and 𝑒 relatively prime positive integers. The subquotient
category of 𝒪𝑐(𝑆𝑛, h) obtained as the quotient of the full subcategory of modules in
𝒪𝑐(𝑆𝑛, h) supported on the subvariety 𝑆𝑛h𝑆𝑘𝑒 , where 𝑘 ≥ 0 is a nonnegative integer,
modulo the Serre subcategory of modules with strictly smaller support, is equivalent to
the category of finite-dimensional modules over the algebra C[𝑆𝑘] ⊗ H𝑞(𝑆𝑛−𝑘𝑒), where
H𝑞(𝑆𝑛−𝑘𝑒) is the Hecke algebra of 𝑆𝑛−𝑘𝑒 with parameter 𝑞 = 𝑒−2𝜋𝑖𝑐.
82
Let 𝑐 = 𝑟/𝑒 be as in Theorem 3.3.3.1. By [3, Theorem 1.2], the only parabolic sub-
groups 𝑊 ′ of 𝑆𝑛 such that 𝐻𝑐(𝑊′, h𝑊 ′) has nonzero finite-dimensional representations
are the conjugates of parabolic subgroups of the form 𝑆𝑘𝑒 for some nonnegative integer
𝑘 ≤ 𝑛/𝑒, and the unique irreducible finite-dimensional representation of 𝐻𝑐(𝑆𝑘𝑒 , h𝑆𝑘
𝑒),
up to isomorphism, is 𝐿 := 𝐿(triv), where triv denotes the trivial representation of
𝑆𝑘𝑒 . It follows that the subquotient category appearing in Theorem 3.3.3.1 is the
subquotient 𝒪𝑐,𝑆𝑘𝑒 ,𝐿
. By Theorem 3.2.2.4, to prove Theorem 3.3.3.1 it suffices to give
an isomorphism of algebras
ℋ(𝑐, 𝑆𝑘𝑒 , 𝐿(triv), 𝑆𝑛) ∼= C[𝑆𝑘] ⊗ H𝑞(𝑆𝑛−𝑒𝑘).
This follows from Theorems 3.3.1.1 and 3.3.2.11, as follows.
The fixed space h𝑆𝑘𝑒 is
h𝑆𝑘𝑒 = (𝑧1, ..., 𝑧𝑛) ∈ h : for 0 ≤ 𝑙 < 𝑘, 𝑧𝑙𝑒+𝑖 = 𝑧𝑙𝑒+𝑗 for 1 ≤ 𝑖 < 𝑖 ≤ 𝑒.
Take coordinates 𝑥1, ..., 𝑥𝑘, 𝑦1, ..., 𝑦𝑛−𝑒𝑘 for h𝑆𝑘𝑒 , where 𝑥𝑙 = 𝑧(𝑙−1)𝑒+𝑖 for 1 ≤ 𝑙 ≤ 𝑘
and 1 ≤ 𝑖 ≤ 𝑒 and 𝑦𝑗 = 𝑧𝑒𝑘+𝑗 for 1 ≤ 𝑗 ≤ 𝑛 − 𝑒𝑘. These coordinates satisfy
the relation∑
𝑖 𝑥𝑖 + 𝑒∑
𝑗 𝑦𝑗 = 0. The complement 𝑁𝑆𝑘𝑒
to 𝑆𝑘𝑒 in its normalizer is
isomorphic to 𝑆𝑘 × 𝑆𝑛−𝑒𝑘, with the action of 𝑆𝑘 on h𝑆𝑘𝑒 given by permuting the 𝑥𝑖
coordinates and the action of 𝑆𝑛−𝑒𝑘 given by permuting the 𝑦𝑗 coordinates. The action
on 𝐻𝑐(𝑆𝑘𝑒 , h𝑆𝑘
𝑒) ∼= 𝐻𝑐(𝑆𝑒, h𝑆𝑒)
⊗𝑘 is by permuting the tensor factors, and in particular
the inertia group 𝐼𝑆𝑘𝑒 ,𝐿
is maximal, i.e. equals 𝑁𝑆𝑘𝑒. Clearly the action of 𝑁𝑆𝑘
𝑒on h𝑆
𝑘𝑒
is generated by reflections, so we have 𝐼𝑟𝑒𝑓𝑆𝑘𝑒 ,𝐿
= 𝑁𝑆𝑘𝑒
and 𝐼𝑐𝑜𝑚𝑝𝑆𝑘𝑒 ,𝐿
= 1. The trivial action
of 𝑁𝑆𝑘𝑒
on the trivial representation triv of 𝑆𝑘𝑒 makes triv equivariant. In particular,
by Remark 3.2.2.3, the 2-cocycle 𝜇 ∈ 𝑍2(𝐼𝑆𝑘𝑒,C×) is trivial.
There are three distinct 𝑁𝑆𝑘𝑒-orbits of hyperplanes defining h
𝑆𝑘𝑒𝑟𝑒𝑔 ⊂ h𝑆
𝑘𝑒 , given by
(1) 𝑥𝑖 = 𝑥𝑗 for 1 ≤ 𝑖 < 𝑗 ≤ 𝑘, (2) 𝑥𝑖 = 𝑦𝑗 for 1 ≤ 𝑖 ≤ 𝑘 and 1 ≤ 𝑗 ≤ 𝑛 − 𝑒𝑘, and
(3) 𝑦𝑖 = 𝑦𝑗 for 1 ≤ 𝑖 < 𝑗 ≤ 𝑛 − 𝑒𝑘. The stabilizers in 𝑆𝑛 of the 𝑥𝑖 = 𝑦𝑗 hyperplanes
are those parabolic subgroups containing 𝑆𝑘𝑒 and conjugate to 𝑆𝑘−1𝑒 × 𝑆𝑒+1, and 𝑆𝑘𝑒
is self-normalizing in these groups. The stabilizers of the hyperplanes 𝑥𝑖 = 𝑥𝑗 are
83
those parabolic subgroups of 𝑆𝑛 containing 𝑆𝑘𝑒 and conjugate to 𝑆2𝑒 × 𝑆𝑘−2𝑒 , and the
stabilizers of the hyperplanes 𝑦𝑖 = 𝑦𝑗 are of those parabolic subgroups of 𝑆𝑛 containing
𝑆𝑘𝑒 and conjugate to 𝑆𝑘𝑒 × 𝑆2. Note that 𝑆𝑘𝑒 is not self-normalizing in either of these
types of parabolic subgroups, and in particular the space h𝑆𝑘𝑒𝐿−𝑟𝑒𝑔 is the complement
of the hyperplanes 𝑥𝑖 = 𝑥𝑗 and 𝑦𝑖 = 𝑦𝑗. It follows already from Theorem 3.3.1.1 that
there is an isomorphism of algebras ℋ(𝑐, 𝑆𝑘𝑒 , 𝐿(triv), 𝑆𝑛) ∼= H𝑞1(𝑆𝑘) ⊗ H𝑞2(𝑆𝑛−𝑒𝑘), for
some complex parameters 𝑞1, 𝑞2 ∈ C×. To show Theorem 3.3.3.1, it therefore suffices
to show that 𝑞1 = 1 and 𝑞2 = 𝑒−2𝜋𝑖𝑐.
By Remark 3.2.2.8, the parameter 𝑞1 can be computed by studying the inclusion
𝑆2𝑒 ⊂ 𝑆2𝑒 and the parameter 𝑞2 can be computed by studying the inclusion 1 ⊂ 𝑆2. In
the latter case, the associated central element 𝑧𝑇1 ∈ H𝑞(1) = C is 1−𝑞 = 1−𝑒−2𝜋𝑖𝑐, and
therefore by Theorem 3.3.2.11 the associated quadratic relation is 𝑇 2 = (1− 𝑞)𝑇 + 𝑞,
so 𝑞2 = 𝑞 = 𝑒−2𝜋𝑖𝑐.
To obtain the parameter 𝑞1, we need to analyze the inclusion 𝑆2𝑒 ⊂ 𝑆2𝑒 and the
associated element 𝑧𝑤0𝑤′0∈ H𝑞(𝑆
2𝑒 ), where 𝑤0 is the longest element of 𝑆2𝑒 and 𝑤′
0 is
the longest element of 𝑆2𝑒 .
Proposition 3.3.3.2. The decomposition of the element 𝑇 2𝑤0𝑤′
0in the 𝑇𝑤-basis of
H𝑞(𝑆2𝑒) is given by
𝑇 2𝑤0𝑤′
0=∑𝑤∈𝑋𝑒
(1 − 𝑞)𝑎(𝑤)𝑞𝑏(𝑤)𝑇𝑤
where 𝑋𝑒 ⊂ 𝑆2𝑒 is the subset of elements 𝑤 ∈ 𝑆2𝑒 such that the three conditions
(1) 𝑤2 = 1
(2) 𝑤(𝑖) = 𝑖 or 𝑤(𝑖) > 𝑒 for 1 ≤ 𝑖 ≤ 𝑒
(3) 𝑤(𝑖) = 𝑖 or 𝑤(𝑖) ≤ 𝑒 for 𝑚 < 𝑖 ≤ 2𝑒,
hold and where the functions 𝑎, 𝑏 : 𝑋𝑒 → Z≥0 are defined by
𝑎(𝑤) = #𝑖 ∈ [1, 𝑒] : 𝑤(𝑖) > 𝑒
84
and
𝑏(𝑤) = −#(𝑖, 𝑗) : 1 ≤ 𝑖 < 𝑗 ≤ 𝑒, 𝑤(𝑖) > 𝑤(𝑗) +𝑒∑𝑖=1
⎧⎪⎨⎪⎩𝑒 𝑤(𝑖) = 𝑖
2𝑒− 𝑤(𝑖) 𝑤(𝑖) > 𝑒.
In particular, the element 𝑧𝑤0𝑤′0∈ H𝑞(𝑆
2𝑒 ) = H𝑞(𝑆𝑒)
⊗2 is given by
𝑧𝑤0𝑤′0
= (1 − 𝑞)𝑒𝑞(𝑒2)∑𝑤∈𝑆𝑒
𝑞−𝑙(𝑤)𝑇𝑤 ⊗ 𝑇𝑤−1 .
Proof. The expression for 𝑇 2𝑤0𝑤′
0can be obtained by a simple inductive argument using
the reduced expression
𝑤0𝑤′0 = (𝑠𝑒 · · · 𝑠2𝑒−1)(𝑠𝑒−1 · · · 𝑠2𝑒−2) · · · (𝑠1 · · · 𝑠𝑒)
and the relations defining the Hecke algebra H𝑞(𝑆2𝑒), from which the expression for
𝑧𝑤0𝑤′0
follows immediately.
The image of the Verma module ∆𝑐(triv) in 𝒪𝑐(𝑆2𝑒 , h𝑆2
𝑒) under the 𝐾𝑍 functor is
the 1-dimensional representation 𝐾𝑍(∆𝑐(triv)) on which all of the generators 𝑇𝑖 ∈
H𝑞(𝑆2𝑒 ) act by the identity. In particular, the element 𝑧𝑤0𝑤′
0acts by the scalar (1 −
𝑞)𝑞(𝑒2)∑
𝑤∈𝑆𝑒𝑞−𝑙(𝑤) = (1−𝑞)𝑞(
𝑒2)𝑃𝑆𝑒(𝑞
−1), where 𝑃𝑆𝑒 is the Poincaré polynomial of 𝑆𝑒.
By the well known identity
𝑃𝑆𝑒 =𝑒∏𝑖=1
1 − 𝑞𝑖
1 − 𝑞
and the fact that 𝑞 = 𝑒−2𝜋𝑖𝑐 is a primitive 𝑒𝑡ℎ root of unity, it follows that 𝑧𝑤0𝑤′0
acts
by 0 on 𝐾𝑍(∆𝑐(triv)), and hence it follows that 𝑧𝑤0𝑤′0
acts by 0 on all simple objects
in the block of H𝑞(𝑆2𝑒 ) containing 𝐾𝑍(∆𝑐(triv)). In particular, 𝐶𝑧𝑤0𝑤′
0(𝐿(triv)) = 0.
Note also that 𝑙(𝑤0𝑤′0) = 𝑒2, so 𝑞𝑙(𝑤0𝑤′
0) = 1. By Theorem 3.3.2.11, it follows that
𝑇 2𝑆2𝑒 ,𝐿,𝑆2𝑒
= 1. Therefore, the parameter 𝑞1 is 1, and the isomorphism
ℋ(𝑐, 𝑆𝑘𝑒 , 𝐿, 𝑆𝑛) ∼= C[𝑆𝑘] ⊗ H𝑞(𝑆𝑛−𝑒𝑘)
85
follows, as needed.
3.3.4 Type 𝐵
In this section we will illustrate our results and check Theorem 3.3.2.14 in the setting
of the type 𝐵 Coxeter groups. The results we obtain in type 𝐵 follow from the work
of Shan-Vasserot [66].
Recall that the Coxeter group 𝐵𝑛 is the semidirect product 𝑆𝑛 n (Z/2Z)𝑛, where
𝑆𝑛 acts on (Z/2Z)𝑛 by permutation, and that it acts in its reflection representation
h = C𝑛 by permutations and sign changes of the coordinates. There are two conjugacy
classes of reflections, the first associated to the coordinate hyperplanes 𝑧𝑖 = 0 and
the second associated to the hyperplanes 𝑧𝑖 = ±𝑧𝑗. A class function 𝑐 on the set
of reflections therefore amounts to a choice of parameter 𝑐1 ∈ C for the 𝑧𝑖 = 0
hyperplanes and a choice of parameter 𝑐2 ∈ C for the 𝑧𝑖 = ±𝑧𝑗 hyperplanes. Let
𝐻𝑐(𝐵𝑛, h) be the associated rational Cherednik algebra. We will choose the set of
simple reflections 𝑠0, 𝑠1, ..., 𝑠𝑛−1 so that 𝑠0 is the reflection through the hyperplane
𝑧1 = 0, negating the first coordinate, and so that 𝑠𝑖, for 0 < 𝑖 < 𝑛, is the reflection
through the hyperplane 𝑧𝑖 = 𝑧𝑖+1, transposing the 𝑖𝑡ℎ and (𝑖+ 1)𝑠𝑡 coordinates.
Recall that the irreducible representations of the Coxeter group 𝐵𝑛 are naturally
labeled by pairs of partitions, or bipartitions, 𝜆 = (𝜆1, 𝜆2) ⊢ 𝑛 of 𝑛 (see, for example,
[37]). In particular, the simple objects in 𝒪𝑐(𝐵𝑛, h) are also labeled by bipartitions,
with the bipartition 𝜆 corresponding to the irreducible representation 𝐿(𝜆) := 𝐿(𝑉𝜆),
where 𝑉𝜆 is the associated irreducible representation of 𝐵𝑛 and 𝐿(𝑉𝜆) is the unique
simple quotient of the Verma module ∆𝑐(𝑉𝜆) attached to 𝑉𝜆. The representation
theory of 𝐻𝑐(𝐵𝑛, h) is much richer than that of 𝐻𝑐(𝑆𝑛, h𝑆𝑛), and in particular the
latter algebra may admit many nonisomorphic irreducible finite-dimensional repre-
sentations.
Any parabolic subgroup of 𝐵𝑛 is conjugate to a unique parabolic subgroup of the
form 𝐵𝑙×𝑆𝑛1 × · · · ×𝑆𝑛𝑘for some nonnegative integers 𝑘, 𝑙 ≥ 0 and positive integers
𝑛1 ≥ · · · ≥ 𝑛𝑘 > 0 with 𝑙 +∑
𝑖 𝑛𝑖 ≤ 𝑛. By [3, Theorem 1.2], the only such parabolic
subgroups whose rational Cherednik algebras admit nonzero finite-dimensional rep-
86
resentations are those of the form 𝐵𝑙×𝑆𝑘𝑒 . Let 𝑊 ′ be such a parabolic subgroup, and
let 𝐿 be an irreducible finite-dimensional representation of 𝐻𝑐(𝑊′, h𝑊 ′). Then 𝐿 is of
the form 𝐿(𝑉𝜆⊗ triv⊗𝑘𝑒 ), where 𝜆 is a bipartition of 𝑙, 𝑉𝜆 is the associated irreducible
representation of 𝐵𝑙, and triv𝑒 denotes the trivial representation of 𝑆𝑒. If 𝑘 > 0 and
such a finite-dimensional irreducible representation exists, the parameter 𝑐2 must be
of the form 𝑟/𝑒 for some integer 𝑟 relatively prime to 𝑒.
The fixed space h𝑊′ ⊂ C𝑛 consists of those points (𝑧1, ..., 𝑧𝑛) such that both
𝑧𝑖 = 0 for 1 ≤ 𝑖 ≤ 𝑙 and also for 0 ≤ 𝑚 < 𝑘 we have 𝑧𝑙+1+𝑚𝑒+𝑖 = 𝑧𝑙+1+𝑚𝑒+𝑗
for 1 ≤ 𝑖, 𝑗 ≤ 𝑒. For 1 ≤ 𝑖 ≤ 𝑘 let 𝑥𝑖 denote the coordinate of 𝑧𝑙+1+𝑖𝑒+1 in h𝑊′ ,
and for 1 ≤ 𝑗 ≤ 𝑛 − 𝑙 − 𝑘𝑒 let 𝑦𝑗 denote the coordinate 𝑧𝑗+𝑙+𝑘𝑒, so that h𝑊′ is
identified with C𝑘 ⊕ C𝑛−𝑙−𝑘𝑒 where the 𝑥𝑖 give the standard coordinates for C𝑘 and
the 𝑦𝑗 give the standard coordinates for C𝑛−𝑙−𝑘𝑒. The natural complement 𝑁𝑊 ′ to
𝑊 ′ in its normalizer 𝑁𝐵𝑛(𝑊 ′) is isomorphic to 𝐵𝑘 × 𝐵𝑛−𝑙−𝑒𝑘 compatibly with the
natural reflection representation of the latter group on C𝑘 ⊕ C𝑛−𝑙−𝑘𝑒. In particular,
𝑁𝑊 ′ = 𝑁 𝑟𝑒𝑓𝑊 ′ . Each parabolic subgroup 𝑊 ′′ ⊂ 𝐵𝑛 containing 𝑊 ′ in corank 1 is
conjugate to a unique parabolic subgroup appearing among the five following cases;
the form of the fixed hyperplane h𝑊′′ ⊂ h𝑊
′ is listed after the parabolic subgroup
𝑊 ′′:
(1) 𝐵𝑙+𝑘 × 𝑆𝑘−1𝑒 𝑥𝑖 = 0
(2) 𝐵𝑙 × 𝑆2𝑒 × 𝑆𝑘−2𝑒 𝑥𝑖 = ±𝑥𝑗
(3) 𝐵𝑙+1 × 𝑆𝑘𝑒 𝑦𝑖 = 0
(4) 𝐵𝑙 × 𝑆𝑘𝑒 × 𝑆2 𝑦𝑖 = ±𝑦𝑗
(5) 𝐵𝑙 × 𝑆𝑒+1 × 𝑆𝑘−1𝑒 𝑥𝑖 = ±𝑦𝑗.
The only such parabolic subgroups in which 𝑊 ′ is self-normalizing are those of type
(5). Furthermore, as the longest element of any Coxeter group of type 𝐵 acts by -1
on its reflection representation, it follows that endowing 𝑉𝜆 ⊗ triv⊗𝑘𝑒 with the trivial
representation of 𝑁𝑊 ′ gives equivariant structure to 𝐿(𝑉𝜆 ⊗ triv⊗𝑘𝑒 ). In particular,
𝐼 = 𝑁𝑊 ′ = 𝑁 𝑟𝑒𝑓𝑊 ′ , the cocycle 𝜇 ∈ 𝑍2(𝐼,C×) is trivial, and h𝑊
′𝐿−𝑟𝑒𝑔 is the complement
in h𝑊′ of the hyperplanes of the forms (1) - (4). In particular, by Theorem 3.3.1.1 we
87
have an isomorphism of algebras
ℋ(𝑐, 𝐵𝑙 × 𝑆𝑘𝑒 , 𝐿(𝑉𝜆 ⊗ triv⊗𝑘𝑒 ), 𝐵𝑛) ∼= H𝑞1,𝑞2(𝐵𝑘) ⊗ H𝑞3,𝑞4(𝐵𝑛−𝑙−𝑘𝑒),
where 𝑞𝑖 ∈ C× is the complex number such that the monodromy operator (appro-
priately scaled) associated to the hyperplanes of type (i), as listed above, satisfies
the quadratic relation (𝑇 − 1)(𝑇 + 𝑞𝑖) = 0. Here 𝑞1 and 𝑞3 are associated to the
reflections through hyperplanes 𝑥𝑖 = 0 and 𝑦𝑖 = 0, respectively, and 𝑞2 and 𝑞4 are as-
sociated to the reflections through hyperplanes 𝑥𝑖 = ±𝑥𝑗 and 𝑦𝑖 = ±𝑦𝑗, respectively.
In particular, Theorem 3.3.2.14 holds in type 𝐵.
By Remark 3.2.2.8 the parameter 𝑞2 associated to the inclusion 𝐵𝑙 × 𝑆𝑘𝑒 ⊂ 𝐵𝑙 ×
𝑆2𝑒 × 𝑆𝑘−2𝑒 can be computed using the inclusion 𝑆2
𝑒 ⊂ 𝑆2𝑒 and the finite-dimensional
irreducible representation 𝐿(triv⊗2𝑒 ) of 𝐻𝑐(𝑆
2𝑒 , h𝑆2
𝑒). This case was treated in Section
3.3.3, and we have 𝑞2 = 1. Similarly, 𝑞4 can by computed using the inclusion 1 ⊂ 𝑆2,
where 𝑆2 is generated by a reflection in 𝐵𝑛 associated to a hyperplane 𝑧𝑖 = ±𝑧𝑗 for any
𝑖, 𝑗 > 𝑙+𝑘𝑒, giving 𝑞4 = 𝑒−2𝜋𝑖𝑐2 . The computation of the parameters 𝑞1 and 𝑞3 reduce to
computing the parameters 𝑞(𝑐, 𝐵𝑘×𝑆𝑒, 𝐿(𝑉𝜆⊗ triv𝑒), 𝐵𝑘+𝑒) and 𝑞(𝑐, 𝐵𝑘, 𝐿(𝑉𝜆), 𝐵𝑘+1),
respectively, which in turn can be computed using Theorem 3.3.2.11. The parameter
𝑞1 = 𝑞(𝑐, 𝐵𝑘×𝑆𝑒, 𝐿(𝑉𝜆⊗ triv𝑒), 𝐵𝑘+𝑒) was computed explicitly in [59, Theorem 2.12],
and we have 𝑞1 = −(−𝑒−2𝜋𝑖𝑐1)𝑒, with no dependence on 𝑘 or 𝜆.
We now explain how to compute the parameter 𝑞3 = 𝑞(𝑐, 𝐵𝑛, 𝐿(𝑉𝜆), 𝐵𝑛+1) from
only the parameter 𝑐 and the bipartition 𝜆 ⊢ 𝑛. Let 𝑝 = 𝑒−2𝜋𝑖𝑐1 and 𝑞 = 𝑒−2𝜋𝑖𝑐2 ,
so that H𝑝,𝑞(𝐵𝑛) is the Hecke algebra appearing in the 𝐾𝑍 functor for 𝒪𝑐(𝐵𝑛, h𝐵𝑛),
where the parameter 𝑝 is associated with reflections through hyperplanes 𝑧𝑖 = 0 and 𝑞
is associated with reflections through hyperplanes 𝑧𝑖 = ±𝑧𝑗. Let 𝑤0 denote the longest
element of 𝐵𝑛+1, let 𝑤′0 denote the longest element of 𝐵𝑛, and let 𝑧𝑤0𝑤′
0denote the
associated central element in H𝑝,𝑞(𝐵𝑛). For 1 ≤ 𝑖 ≤ 𝑛, let 𝑡𝑖 ∈ 𝐵𝑛 denote the reflection
𝑡𝑖 := 𝑠𝑖−1 · · · 𝑠1𝑠0𝑠1 · · · 𝑠𝑖−1 negating the 𝑖𝑡ℎ coordinate.
Fix a bipartition 𝜆 = (𝜆(1), 𝜆(2)) ⊢ 𝑛 of 𝑛, and for 𝑖 = 1, 2 let 𝜆(𝑖)1 ≥ · · · ≥ 𝜆(𝑖)𝑙𝑖> 0
be the parts of the partition 𝜆(𝑖). Recall that we may view 𝜆 as a pair of Young
88
diagrams in the following way. Refer to an element 𝑏 = (𝑥, 𝑦, 𝑖) ∈ Z>0 × Z>0 × 1, 2
as a box. A finite subset 𝑌 ⊂ Z>0×Z>0×1, 2 is called a Young diagram if whenever
𝑌 contains the box (𝑥, 𝑦, 𝑖) it also contains all boxes of the form (𝑥′, 𝑦′, 𝑖) for positive
integers 𝑥′, 𝑦′ satisfying 1 ≤ 𝑥′ ≤ 𝑥 and 1 ≤ 𝑦′ ≤ 𝑦. Let 𝑌 𝐷(𝜆) ⊂ Z>0 × Z>0 × 1, 2
be the Young diagram consisting of those boxes (𝑥, 𝑦, 𝑖) such that 𝑦 ≤ 𝑙𝑖 and 𝑥 ≤ 𝜆(𝑖)𝑦 .
Define the content, with respect to the parameters 𝑝 and 𝑞, of a box 𝑏 = (𝑥, 𝑦, 𝑖) to
be 𝑞𝑥−𝑦𝑝−1 if 𝑖 = 1 and to be −𝑞𝑥−𝑦 if 𝑖 = 2. Denote the content of 𝑏 by 𝑐𝑡𝑝,𝑞(𝑏).
Definition 3.3.4.1. Given a bipartition 𝜆 ⊢ 𝑛 of 𝑛 and parameters 𝑝, 𝑞 ∈ C× for the
Hecke algebra H𝑝,𝑞(𝐵𝑛), define the scalar 𝑧𝑝,𝑞(𝜆) ∈ C by
𝑧𝑝,𝑞(𝜆) := (1 − 𝑝)𝑞𝑛 + (1 − 𝑞)2𝑞𝑛−1𝑝∑
𝑏∈𝑌 𝐷(𝜆)
𝑐𝑡𝑝,𝑞(𝑏).
Remark 3.3.4.2. Our definition of content differs slightly from the definition of
content appearing in [37, Section 10.1.4] because we choose a different convention for
the quadratic relations satisfied by the generators 𝑇𝑖 of H𝑝,𝑞(𝐵𝑛), i.e. that the quadratic
relations should be divisible by (𝑇 − 1) rather than (𝑇 + 1). This is natural from the
perspective of the 𝐾𝑍 functor. Our algebra H𝑝,𝑞(𝐵𝑛) is isomorphic to the algebra
H𝑝−1,𝑞−1(𝐵𝑛) appearing in [37] under the isomorphism induced by the assignments
𝑇𝑠 ↦→ 𝑞−1𝑠 𝑇𝑠. The inversion of the parameters explains the discrepancy in the definition
of content.
Proposition 3.3.4.3. In the notation of Lemma 3.3.2.7, 𝑞𝑙(𝑤0𝑤′0) = 𝑝𝑞2𝑛, and the
central element 𝑧𝑤0𝑤′0∈ H𝑝,𝑞(𝐵𝑛) has the following expansion in the 𝑇𝑤-basis:
𝑧𝑤0𝑤′0
= (1 − 𝑝)𝑞𝑛 + (1 − 𝑞)2𝑛∑𝑖=1
𝑞𝑛−𝑖𝑇𝑡𝑖 .
Moreover, 𝑧𝑤0𝑤′0
acts on any irreducible representation of H𝑝,𝑞(𝐵𝑛) lying in the block
of H𝑝,𝑞(𝐵𝑛)-mod𝑓.𝑑. corresponding via the 𝐾𝑍 functor to the block of 𝒪𝑐(𝐵𝑛, h𝐵𝑛)
containing 𝐿(𝑉𝜆) by the scalar 𝑧𝑝,𝑞(𝜆) defined in Definition 3.3.4.1.
Proof. The expressions for 𝑧𝑤0𝑤′0
and 𝑞𝑙(𝑤0𝑤′0) follow immediately from standard calcu-
89
lations in H𝑝,𝑞(𝐵𝑛+1) using the reduced expression 𝑠𝑛 · · · 𝑠1𝑠0𝑠1 · · · 𝑠𝑛 for 𝑤0𝑤′0, where
𝑠0 is the simple reflection through the hyperplane 𝑧1 = 0 and, for 𝑖 > 0, 𝑠𝑖 is the sim-
ple reflection through the hyperplane 𝑧𝑖 = 𝑧𝑖+1. To show that 𝑧𝑤0𝑤′0
acts by 𝑧𝑝,𝑞(𝜆) on
the irreducibles in the block of H𝑝,𝑞(𝐵𝑛)-mod𝑓.𝑑. appearing in the theorem, it suffices
to show that 𝑧𝑤0𝑤′0
acts by 𝑧𝑝,𝑞(𝜆) on 𝐾𝑍(∆𝑐(𝑉𝜆)). By a standard deformation argu-
ment, it suffices to prove this for generic parameters 𝑝, 𝑞, 𝑐. For generic parameters,
𝐾𝑍(∆𝑐(𝑉𝜆)) is isomorphic to the irreducible representation 𝑉 𝑝,𝑞𝜆 of H𝑝,𝑞(𝐵𝑛) described
in [37, Theorem 10.1.5] in terms of Hoefsmit’s matrices, and that 𝑧𝑤0𝑤′0
acts by the
scalar 𝑧𝑝,𝑞(𝜆) on 𝑉 𝑝,𝑞𝜆 then follows immediately from the explicit description of the
diagonal action of the elements 𝑇𝑡𝑖 on the standard Young tableau basis of 𝑉 𝑝,𝑞𝜆 .
In particular, by Theorem 3.3.2.11, the canonical generator 𝑇𝐵𝑛,𝐿(𝑉𝜆),𝐵𝑛+1 of the
algebra ℋ(𝑐, 𝐵𝑛, 𝐿(𝑉𝜆), 𝐵𝑛+1) satisfies the same quadratic relation as the matrix
⎛⎝0 𝑝𝑞2𝑛
1 𝑧𝑝,𝑞(𝜆)
⎞⎠ ,
i.e.
𝑇 2 = 𝑧𝑝,𝑞(𝜆)𝑇 + 𝑝𝑞2𝑛.
Rescaling appropriately, one obtains the parameter 𝑞(𝑐, 𝐵𝑛, 𝐿(𝑉𝜆), 𝐵𝑛+1). Note that
when 𝑞 is a primitive 𝑒𝑡ℎ root of unity, this quadratic relation depends only on the
𝑒-cores of the components of 𝜆.
3.3.5 Type 𝐷
In this section we will show that Theorem 3.3.2.14 holds in type 𝐷 and that the study
of the generalized Hecke algebras ℋ(𝑐,𝑊 ′, 𝐿,𝑊 ) when 𝑊 is of type 𝐷 largely reduces
to the case in which 𝑊 is of type 𝐵.
Recall that for 𝑛 ≥ 4 the reflection group 𝐷𝑛 of type 𝐷 and rank 𝑛 is the subgroup
of 𝐵𝑛 of index 2 consisting of those elements acting on C𝑛 with an even number of sign
changes. 𝐷𝑛 is an irreducible reflection group with reflection representation C𝑛 in this
way, generated by reflections through the hyperplanes 𝑧𝑖 = ±𝑧𝑗 for 1 ≤ 𝑖 < 𝑗 ≤ 𝑛.
90
If 𝑠0, ..., 𝑠𝑛−1 are the simple reflections for 𝐵𝑛 introduced in the previous section,
then the reflections 𝑠′1, 𝑠1, 𝑠2, ..., 𝑠𝑛−1, where 𝑠′1 = 𝑠0𝑠1𝑠0 is the reflection through the
hyperplane 𝑧1 = −𝑧2, form a system of simple reflections for 𝐷𝑛 with respect to which
it is a Coxeter group.
The irreducible complex representations of 𝐷𝑛 are easily described in terms of
those of 𝐵𝑛 recalled in the previous section. In particular, when 𝜆 = (𝜆1, 𝜆2) is a
bipartition of 𝑛 for which 𝜆1 = 𝜆2, then the restriction of 𝑉𝜆 to 𝐷𝑛 is irreducible,
and 𝑉(𝜆1,𝜆2) and 𝑉(𝜆2,𝜆1) are isomorphic as representations of 𝐷𝑛. When 𝜆1 = 𝜆2, i.e.
when the bipartition 𝜆 is symmetric, the restriction of 𝑉𝜆 to 𝐷𝑛 splits as a direct
sum of two non-isomorphic irreducible representations 𝑉 +𝜆 and 𝑉 −
𝜆 . All irreducible
representations of 𝐷𝑛 appear in this way, and the only isomorphisms among these
representations are those of the form 𝑉(𝜆1,𝜆2) ∼= 𝑉(𝜆2,𝜆1).
All reflections in 𝐷𝑛 are conjugate, so a parameter for the rational Cherednik
algebra of type 𝐷 is determined by a single number 𝑐 ∈ C. It follows immediately
from the definition by generators and relations that the rational Cherednik algebra
𝐻𝑐(𝐷𝑛,C𝑛) embeds naturally in the type 𝐵 algebra 𝐻0,𝑐(𝐵𝑛,C𝑛), where in the latter
algebra the parameter takes value 0 on reflections through hyperplanes 𝑧𝑖 = 0 and
value 𝑐 on reflections through hyperplanes 𝑧𝑖 = ±𝑧𝑗. Let 𝑞 = 𝑒−2𝜋𝑖𝑐 be the parameter
for the Hecke algebra H𝑞(𝐷𝑛) whose category of finite-dimensional modules is the
target of the 𝐾𝑍 functor. Note similarly that the Hecke algebra H𝑞(𝐷𝑛) embeds
naturally as a subalgebra of the Hecke algebra 𝐻1,𝑞(𝐵𝑛) compatibly with the 𝑇𝑤 bases
(see [37, Section 10.4.1]); note that 𝑇 2𝑠0
= 1. This embedding is compatible with the
𝐾𝑍 functors in the obvious way. It is shown in [68] that when the bipartition 𝜆
is symmetric the irreducible representations 𝐿𝑐(𝑉 ±𝜆 ) are always infinite dimensional.
In particular, the finite-dimensional irreducible representations of 𝐻𝑐(𝐷𝑛,C𝑛) always
extend to irreducible representations of 𝐻0,𝑐(𝐵𝑛,C𝑛), although not uniquely.
Suppose 𝑊 ′ ⊂ 𝐷𝑛 is a standard parabolic subgroup such that 𝐻𝑐(𝑊′, h𝑊 ′) admits
a finite-dimensional irreducible representation 𝐿. The irreducible parabolic subgroups
of 𝐷𝑛 are of types 𝐴 and 𝐷. We will now describe a procedure for producing a
presentation of the algebra ℋ(𝑐,𝑊 ′, 𝐿,𝐷𝑛) in the form appearing in Theorem 3.3.2.14,
91
and in particular we will see that Theorem 3.3.2.14 holds in type 𝐷. Clearly, by
tensoring with the sign character of 𝐷𝑛, we may assume 𝑐 > 0.
First suppose that the decomposition of 𝑊 ′ into a product of irreducible parabolic
subgroups involves no factors of type𝐷. Then by [3, Theorem 1.2] we can assume that
𝑐 = 𝑟/𝑒 for positive relatively prime integers 𝑟 ≥ 1, 𝑒 ≥ 2, that 𝑊 ′ is of the form 𝑆𝑘𝑒 for
some integer 𝑘 > 0 such that 𝑘𝑒 ≤ 𝑛, and that 𝐿 = 𝐿(triv⊗𝑘𝑒 ). It follows from Remark
3.2.2.3 that the inertia group 𝐼 equals the complement 𝑁𝑊 ′ to 𝑊 ′ in 𝑁𝐷𝑛(𝑊 ′) and
that the cocycle 𝜇 ∈ 𝑍2(𝑁𝑊 ′ ,C×) is trivial. In particular, Theorem 3.3.2.14 holds
in this case. A detailed description of the group 𝑁𝑊 ′ and its maximal reflection
subgroup 𝑁 𝑟𝑒𝑓𝑊 ′ (typically a proper subgroup of 𝑁𝑊 ′) may be found in [47]. As usual,
the parameter 𝑞𝑊 ′,𝐿 associated to the Hecke algebra H𝑞𝑊 ′,𝐿(𝑁 𝑟𝑒𝑓
𝑊 ′ ) can be computed
using Theorem 3.3.2.11. The parabolic subgroups 𝑊 ′′ ⊂ 𝐷𝑛 containing 𝑊 ′ in corank
1 and in which 𝑊 ′ is not self-normalizing are of the form (1) 𝑆𝑘𝑒 ×𝑆2, (2) 𝑆𝑘−2𝑒 ×𝑆2𝑒,
(3) (in the case 𝑒 = 2) 𝑆𝑘−32 ×𝐷4, and (4) (in the case 𝑒 = 4) 𝑆𝑘−1
4 ×𝐷4. By Remark
3.2.2.8, parameter computations in these cases reduce to the cases, respectively, (1)
1 ⊂ 𝑆2, (2) 𝑆2𝑒 ⊂ 𝑆2𝑒, (3) 𝑆3
2 ⊂ 𝐷4, and (4) 𝑆4 ⊂ 𝐷4. As discussed in Section 3.3.3
about type 𝐴, the quadratic relation in case (1) is (𝑇 −1)(𝑇 + 𝑞) = 0 with 𝑞 = 𝑒−2𝜋𝑖𝑐,
and the quadratic relation in case (2) is 𝑇 2 = 1. To compute the quadratic relation in
case (3), we use Theorem 3.3.2.11 again. In particular, letting 𝑤0 denote the longest
element of 𝐷4 and 𝑤′0 the longest element of 𝑆3
2 , we have 𝑞𝑙(𝑤0𝑤′0) = (−1)9 = −1, and
computations in the computer algebra package CHEVIE in GAP3 [36, 58] show that
the central element 𝑧𝑤0𝑤′0
acts on the trivial representation of 𝐻−1(𝑆2)⊗3 by the scalar
2. By Theorem 3.3.2.11 the quadratic relation appearing in case (3) is 𝑇 2 = 2𝑇 − 1,
i.e. (𝑇 − 1)2 = 0. Similarly, one obtains the quadratic relation (𝑇 − 1)2 = 0 in
remaining case (4) as well.
Now, consider the remaining case in which the decomposition of 𝑊 ′ into a prod-
uct of irreducible parabolic subgroups involves a factor of type 𝐷. Then again by
[3, Theorem 1.2] we can assume that 𝑊 ′ is conjugate to a parabolic subgroup of
the form 𝐷𝑙 × 𝑆𝑘𝑒 for some integers 𝑒 ≥ 2, 𝑙 ≥ 4, 𝑘 ≥ 0 such that 𝑙 + 𝑘𝑒 ≤ 𝑛, and
the finite-dimensional irreducible representation 𝐿 is isomorphic to 𝐿𝑐(𝑉𝜆 ⊗ triv⊗𝑘𝑒 )
92
for some bipartition 𝜆 = (𝜆1, 𝜆2) with 𝜆1 = 𝜆2. The parameter 𝑐 must again be of
the form 𝑐 = 𝑟/𝑒 for some positive integer 𝑟 relatively prime to 𝑒. Furthermore, it
follows from [55, Lemma 4.2, Corollary 4.3] that 𝐻𝑐(𝐷𝑙,C𝑙) admits no nonzero finite-
dimensional representations when 𝑒 is odd, so we may assume that 𝑒 is even. To see
this, consider a parameter 𝑐 = 𝑟/𝑒 for the rational Cherednik algebra 𝐻𝑐(𝐷𝑛,C𝑛),
where 𝑒 > 2 is an odd positive integer and 𝑟 an integer relatively prime to 𝑒. The
algebra 𝐻𝑐(𝐷𝑛,C𝑛) admits nonzero finite-dimensional representations if and only if
the algebra 𝐻0,𝑐(𝐵𝑛,C𝑛) admits nonzero finite-dimensional representations. In the
notation from [55, Section 4], in this case we have 𝜅 = −𝑟/𝑒 and (𝑠1, 𝑠2) = (0,− 𝑒2𝑟
).
Indexes 1 and 2 are not equivalent under the equivalence relation ∼(0,𝑐) as we have
𝑠2 − 𝑠1 = − 𝑒2𝑟/∈ 1
𝑟Z = 𝜅−1Z + Z, so by [55, Lemma 4.2] the category 𝒪(0,𝑐)(𝐵𝑛,C𝑛)
decomposes as a direct sum of outer tensor products of categories 𝒪 associated to
reflection groups 𝑆𝑘 with reflection representation C𝑘, for various 𝑘, in a manner pre-
serving supports [55, Corollary 4.3]. As the rational Cherednik algebras associated
to the reducible reflection representations (𝑆𝑘,C𝑘) have no nonzero finite-dimensional
representations for any parameter values, it follows that the rational Cherednik al-
gebra 𝐻𝑟/𝑒(𝐷𝑛,C𝑛) also has no nonzero finite-dimensional representations. We will
therefore assume that 𝑒 > 1 is a positive even integer.
As the fixed space of 𝑊 ′ equals the fixed space of the parabolic subgroup 𝐵𝑙×𝑆𝑘𝑒
of 𝐵𝑛 ⊃ 𝐷𝑛, it follows that 𝑁𝐷𝑛(𝑊 ′) = 𝐷𝑛 ∩ 𝑁𝐵𝑛(𝐵𝑙 × 𝑆𝑘𝑒 ). As 𝜆1 = 𝜆2, the
representation 𝑉𝜆⊗triv⊗𝑘𝑒 of 𝑊 ′ extends to a representation of 𝐵𝑙×𝑆𝑘𝑒 , and we’ve seen
in Section 3.3.4 that such a representation extends to a representation of𝑁𝐵𝑛(𝐵𝑙×𝑆𝑘𝑒 ).
In particular, it follows that the inertia group 𝐼 is maximal, i.e. equals 𝑁𝑊 ′ , and that
the cocycle 𝜇 ∈ 𝑍2(𝑁𝑊 ′ ,C×) is trivial, so Theorem 3.3.2.14 holds in this remaining
case in type 𝐷. As discussed in “Case 1” of the section “Type D” of [47], in this case
𝑁𝑊 ′ equals 𝑁 𝑟𝑒𝑓𝑊 ′ and is isomorphic to 𝐵𝑘 × 𝐵𝑛−𝑙−𝑘𝑒 as a reflection group acting on
(C𝑛)𝑊′ ∼= C𝑘⊕C𝑛−𝑙−𝑘𝑒 in a manner completely analogous to the discussion in Section
3.3.4. In particular, in this case we have
ℋ(𝑐,𝐷𝑙 × 𝑆𝑘𝑒 , 𝐿𝑐(𝑉𝜆 ⊗ triv⊗𝑘𝑒 ), 𝐷𝑛) ∼= H𝑞1,1(𝐵𝑘) ⊗ H𝑞2,𝑞(𝐵𝑛−𝑙−𝑘𝑒)
93
where 𝑞1 and 𝑞2 are associated to the short roots of 𝐵𝑘 and 𝐵𝑛−𝑙−𝑘𝑒, respectively,
𝑞 = 𝑒−2𝜋𝑖𝑐, 𝑞1 = 𝑞(𝑐,𝐷𝑙 × 𝑆𝑒, 𝐿𝑐(𝑉𝜆 ⊗ triv𝑒), 𝐷𝑙+𝑒), and 𝑞2 = 𝑞(𝑐,𝐷𝑙, 𝐿𝑐(𝑉𝜆), 𝐷𝑙+1).
The following result reduces the computation of these parameters to the type 𝐵:
Proposition 3.3.5.1. In the setting of the previous paragraph, we have
𝑞(𝑐,𝐷𝑛, 𝐿𝑐(𝑉𝜆), 𝐷𝑛+1) = 𝑞((0, 𝑐), 𝐵𝑛, 𝐿(0,𝑐)(𝑉𝜆), 𝐵𝑛+1)
and
𝑞(𝑐,𝐷𝑛 × 𝑆𝑒, 𝐿𝑐(𝑉𝜆 ⊗ triv𝑒), 𝐷𝑛+𝑒) = 𝑞((0, 𝑐), 𝐵𝑛 × 𝑆𝑒, 𝐿(0,𝑐)(𝑉𝜆 ⊗ triv𝑒), 𝐵𝑛+𝑒).
Proof. We consider 𝑞(𝑐,𝐷𝑛, 𝐿𝑐(𝑉𝜆), 𝐷𝑛+1) first. Let 𝑙𝐷 denote the length function on
𝐷𝑛+1 with respect to the simple reflections 𝑠1, 𝑠′1, ..., 𝑠𝑛 introduced above, and let 𝑙𝐵
denote the length function on 𝐵𝑛+1 with respect to the simple reflections 𝑠0, 𝑠1, ..., 𝑠𝑛.
Let 𝑤0,𝐷, 𝑤0,𝐵, 𝑤′0,𝐷, and 𝑤′
0,𝐵 denote the longest elements of the Coxeter groups
𝐷𝑛+1, 𝐵𝑛+1, 𝐷𝑛 and 𝐵𝑛, respectively. Then 𝑤0,𝐷𝑤′0,𝐷 = 𝑠0𝑤0,𝐵𝑤
′0,𝐵 = 𝑤0,𝐵𝑤
′0,𝐵𝑠0. Re-
gard H𝑞(𝐷𝑛) as a subalgebra of 𝐻1,𝑞(𝐵𝑛) via the 𝑇𝑤-bases. We then have 𝑇𝑤0,𝐷𝑤′0,𝐷
=
𝑇𝑠0𝑇𝑤0,𝐵𝑤′0,𝐵
= 𝑇𝑤0,𝐵𝑤′0,𝐵𝑇𝑠0 and 𝑇 2
𝑤′0
= 1. In particular, 𝑇 2𝑤0,𝐷𝑤
′0,𝐷
= 𝑇 2𝑤0,𝐵𝑤
′0,𝐵
and it
follows from Proposition 3.3.4.3 and the definitions that 𝑧𝑤0,𝐷𝑤′0,𝐷
= 𝑇𝑠0𝑧𝑤0,𝐵𝑤′0,𝐵
=
𝑧𝑤0,𝐵𝑤′0,𝐵𝑇𝑠0 . As the representation 𝑉𝜆 extends to 𝐵𝑛, we can choose the operator by
which 𝑤0,𝐵𝑤′0,𝐵 acts on 𝐿 as in Theorem 3.3.2.11 to be 𝑇𝑠0 . By Theorem 3.3.2.11, the
generator of monodromy 𝑇 ∈ ℋ(𝑐,𝐷𝑛, 𝐿𝑐(𝑉𝜆), 𝐷𝑛+1) satisfies the quadratic relation
𝑇 2 = (𝑇0𝑧𝑤0,𝐷𝑤′0,𝐷
)|𝐿𝑇 + 𝑞𝑙𝐷(𝑤0,𝐷𝑤′
0,𝐷)
= 𝑧1,𝑞(𝜆)𝑇 + 𝑞2𝑛.
This is precisely the quadratic relation obtained for the generator of monodromy
generating the algebra 𝑇 ∈ ℋ(𝑐,𝐷𝑛, 𝐿𝑐(𝑉𝜆), 𝐷𝑛+1), as shown in Section 3.3.4, and
the first equality follows.
The second equality follows by a similar argument.
94
3.3.6 Parameters for Generalized Hecke Algebras in Excep-
tional Types
We will now describe the parameters arising for the generalized Hecke algebras in
exceptional type. In each row of the following Table 3.1, 𝑊 is an irreducible finite
Coxeter group and 𝑊 ′ ⊂ 𝑊 is a corank-1 parabolic subgroup of 𝑊 that is not self-
normalizing in 𝑊 . The complex number 𝑐 is a parameter for the rational Cherednik
algebra 𝐻𝑐(𝑊, h) such that 𝐻𝑐(𝑊′, h𝑊 ′) admits nontrivial finite-dimensional repre-
sentations, and 𝜆 is an irreducible representation of 𝑊 ′ such that 𝐿(𝜆) is a finite-
dimensional irreducible representation of 𝐻𝑐(𝑊′, h𝑊 ′). For a given 𝑊 ′, all 𝑐 of the
form 1/𝑑 such that 𝐻𝑐(𝑊′, h𝑊 ′) admits a finite-dimensional irreducible representation
are given, and for each 𝑊 ′, 𝑐 a complete list of lowest weights 𝜆 with dim𝐿(𝜆) < ∞
is given. Finally, 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ) is a complex number such that the monodromy op-
erator 𝑇 associated to the tuple (𝑐,𝑊 ′, 𝐿(𝜆),𝑊 ), after an appropriate rescaling if
necessary, satisfies the quadratic relation
(𝑇 − 1)(𝑇 + 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 )) = 0.
Where appropriate, 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ) is given as a power of the “𝐾𝑍 parameter” 𝑞 =
𝑒−2𝜋𝑖𝑐. Table 3.1 includes every case needed to give presentations for the generalized
Hecke algebras arising in types 𝐸,𝐻 and 𝐼; this data was obtained by using Theorem
3.3.2.11 and computations with the computer algebra package CHEVIE in GAP3
[36, 58] as well as SAGE. Type 𝐹 can be handled by these same methods, although
the description of the relevant parameters 𝑐 and irreducible finite-dimensional repre-
sentations for the parabolic subgroups of types 𝐵 and 𝐶 arising in this case is more
complicated to display in a table. We will give the counts of modules of given support
in 𝒪𝑐(𝐹4, h) in the unequal parameter case later in Section 3.3.9.
In Table 3.1 and below, we will list the parameter for groups 𝐵𝑛 in the form
(𝑐1, 𝑐2) ∈ C2, where 𝑐1 specifies the value of the parameter on the short roots. In
the last row of the table, the parameter (1/2, 𝑐2) for the even dihedral group 𝐼2(2𝑚)
indicates that the parameter takes value 1/2 on those reflections conjugate to the
95
nontrivial element of the chosen parabolic subgroup 𝐴1 and arbitrary value 𝑐2 ∈
C on the remaining parameters. The relevant parameter values and lists of finite-
dimensional irreducible representations for the groups of type 𝐷 are obtained by
a standard reduction to type 𝐵 (as in Section 3.3.5), where these lists are easily
produced using the methods of [55]. The labeling used for irreducible representations
of the exceptional groups is compatible with that appearing in [45]; in particular,
we denote the trivial representation by triv, the reflection representation by 𝑉 (and
its Galois conjugate in type 𝐻 by 𝑉 ), and other representations are denoted in the
form 𝜙𝑥,𝑦 where 𝑥 indicates the dimension of the representation and 𝑦 indicates its 𝑏-
invariant, i.e. the lowest degree in the grading of the coinvariant algebra in which the
representation appears. The labels 𝜙𝑥,𝑦 are compatible with the labels appearing in
the GAP3 computer algebra package. This is a different labeling system than appears
in some standard references, e.g. [37], although it is simple to convert between this
labeling system and others using the tables appearing in [37, Appendix C]. Irreducible
representations of groups of type 𝐷 are labeled by (unordered) pairs of partitions, in
the standard way.
Remark 3.3.6.1. In all cases listed except the case of 𝐸7 at parameter 1/10, the
associated monodromy operator 𝑇 has an eigenvalue equal to 1, and in particular no
rescaling was needed to list the parameter 𝑞(𝑐,𝑊 ′, 𝐿(𝜆),𝑊 ); the monodromy operator
𝑇 associated to irreducible representation 𝐿1/10(𝑉 ) of 𝐻1/10(𝐸7, h) associated to the
inclusion 𝐸7 ⊂ 𝐸8 satisfies the quadratic relation (𝑇+𝑒𝜋𝑖/5)2 = 0, which is of the form
(𝑇 − 1)2 = 0 after rescaling 𝑇 . In the cases in which 𝑇 has an eigenvalue equal to 1,
the parameter 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ) is necessarily equal to 𝑞𝑙(𝑤0𝑤′0), where 𝑞 is the parameter
appearing in the relevant 𝐾𝑍 functor and 𝑞𝑙(𝑤0𝑤′0) is as in Theorem 3.3.2.11, and
this covers all other cases in the table. We remark that there are other cases, not
relevant for the exceptional groups, in which 𝑇 does not have an eigenvalue equal
to 1 before rescaling; for example, 𝐿(−1/6,1/3)(triv) is a finite-dimensional irreducible
representation of 𝐻(−1/6,1/3)(𝐵2) and the quadratic relation associated to the inclusion
𝐵2 ⊂ 𝐵3 is (𝑇 + 𝑝)2 = 0, where 𝑝 = 𝑒−𝜋𝑖/3.
96
Remark 3.3.6.2. In all cases we have computed, the parameter 𝑞(𝑐,𝑊 ′, 𝐿,𝑊 ) de-
pends only on 𝑐,𝑊 ′, and 𝑊 , and notably not on the finite-dimensional irreducible
representation 𝐿. This fact is reflected in Table 3.1, where we list all relevant lowest
weights 𝜆 for each pair (𝑊 ′, 𝑐) in the same row. It would be interesting to have a
conceptual explanation for this fact.
Table 3.1: Parameters for Generalized Hecke Algebras
𝑊 ′ 𝑐 𝜆 𝑊 𝑞(𝑐,𝑊 ′, 𝐿(𝜆),𝑊 )
𝐴𝑛 × 𝐴𝑛 1/(𝑛+ 1) triv 𝐴2𝑛+1 1
𝐴31 1/2 triv 𝐷4 −1
𝐴3 1/4 triv 𝐷4 −1
𝐷4 1/6 triv 𝐷5 𝑞2
𝐷4 1/4 triv 𝐷5 1
𝐷4 1/2 triv, (3, 1) 𝐷5 1
𝐴5 1/6 triv 𝐷6 −1
𝐷5 1/8 triv 𝐷6 𝑞2
𝐷4 × 𝐴1 1/2 triv ⊗ triv, (3, 1) ⊗ triv 𝐷6 −1
𝐴23 1/4 triv 𝐷7 -1
𝐷6 1/10 triv 𝐷7 𝑞2
𝐷6 1/6 triv 𝐷7 1
𝐷6 1/2 triv, (0, 32), (1, 5), (2, 4) 𝐷7 1
𝐴5 1/6 triv 𝐸6 −1
𝐴6 1/7 triv 𝐸7 1
𝐷6 1/10 triv 𝐸7 𝑞3
𝐷6 1/6 triv 𝐸7 −1
𝐷6 1/2 triv, (0, 32), (1, 5), (2, 4) 𝐸7 −1
𝐸6 1/12 triv 𝐸7 𝑞3
𝐸6 1/9 triv 𝐸7 1
𝐸6 1/6 triv, 𝑉 𝐸7 −1
97
Table 3.1: (continued)
𝐸6 1/3 triv, 𝑉,Λ2𝑉 𝐸7 1
𝐴7 1/8 triv 𝐸8 −1
𝐷7 1/12 triv 𝐸8 −1
𝐷7 1/4 triv, (2, 5) 𝐸8 −1
𝐸7 1/18 triv 𝐸8 𝑞3
𝐸7 1/14 triv 𝐸8 𝑞
𝐸7 1/10 𝑉 𝐸8 −1
𝐸7 1/6 triv, 𝑉, 𝜙15,7, 𝜙21,6 𝐸8 −1
𝐸7 1/2 triv, 𝑉, 𝜙15,7, 𝜙21,6
𝜙27,2, 𝜙35,13, 𝜙189,5 𝐸8 −1
𝐴2 (𝑐1, 1/3) triv 𝐵3 𝑒−6𝜋𝑖𝑐1
𝐴21 (1/2, 1/2) triv 𝐵3 −1
𝐴2 1/3 triv 𝐻3 1
𝐴1 × 𝐴1 1/2 triv 𝐻3 −1
𝐼2(5) 1/5 triv 𝐻3 1
𝐻3 1/10 triv 𝐻4 −1
𝐻3 1/6 triv 𝐻4 −1
𝐻3 1/2 triv, 𝑉, 𝑉 𝐻4 −1
𝐴3 1/4 triv 𝐻4 −1
𝐴1 (1/2, 𝑐2) triv 𝐼2(2𝑚) (−1)𝑚−1𝑒−2𝑚𝑐2𝜋𝑖
3.3.7 Type 𝐸
Generalized Hecke Algebras for 𝐸6
The following table list all of the generalized Hecke algebras arising from the rational
Cherednik algebra 𝐻𝑐(𝐸6, h𝐸6) of type 𝐸6 for parameters of the form 𝑐 = 1/𝑑 such
that 𝒪𝑐(𝐸6, h𝐸6) is not semisimple, i.e. for those integers 𝑑 > 1 dividing one of the
fundamental degrees 2, 5, 6, 8, 9 and 12 of 𝐸6. The first column indicates the param-
98
eter value 𝑐. The second column, labeled 𝑊 ′, lists a unique representative of each
conjugacy class of parabolic subgroups of 𝐸6 for which a nonzero finite-dimensional
representation appears at the parameter value specified in the title of the table; if a
conjugacy class is missing, no nonzero finite-dimensional irreducible representations
exist for that class. The column labeled 𝜆 gives a complete list of the lowest weights
𝜆 of the finite-dimensional irreducible representations 𝐿𝑐(𝜆) of 𝐻𝑐(𝑊′, h𝑊 ′). By in-
spection, we see that in each case the inertia group is maximal, the 2-cocycle 𝜇 is
trivial, and in particular Theorem 3.3.2.14 holds for algebras of type 𝐸6. Furthermore,
from Table 3.1 we see that the parameters for the generalized Hecke algebra does not
depend on the choice of lowest weight for the finite-dimensional representation; there-
fore in the column labeled ℋ we give the generalized Hecke algebra ℋ(𝑐,𝑊 ′, 𝜆, 𝐸6)
common to each of the 𝜆 appearing in a given row. In the final column labeled #Irr,
we give the number of irreducible representations with support labeled by 𝑊 ′, ob-
tained as the product of the number of 𝜆 appearing in a given row with the number of
irreducible representations of the corresponding generalized Hecke algebra. As there
are 25 irreducible representations of the group 𝐸6, these numbers add to 25 for each
parameter value.
Throughout, 𝑞 denotes the “𝐾𝑍 parameter” 𝑞 := 𝑒−2𝜋𝑖𝑐. The exact descriptions
of the generalized Hecke algebras follow easily from parameters in Table 3.1 and
Howlett’s detailed descriptions of the groups 𝑁 𝑟𝑒𝑓𝑊 ′ and 𝑁 𝑐𝑜𝑚𝑝
𝑊 ′ appearing in [47]; any
semidirect products appearing in the description of the algebras ℋ are given by the
diagram automorphisms indicated in [47]. By convention, we list the parameters of
2-parameter Hecke algebras by giving the parameter for the short roots first, e.g.
H𝑝,𝑞(𝐵3) indicates that parameter 𝑝 is associated with the 3 reflections given by short
roots and that parameter 𝑞 is associated with the remaining 6 reflections given by
long roots.
The finite-dimensional irreducible representations of 𝐻𝑐(𝐸6, h𝐸6), if they exist,
appear in the rows labeled by 𝐸6. For all parameters except 𝑐 = 1/2, the list of the
lowest weights of the finite-dimensional irreducible representations of 𝐻𝑐(𝐸6, h𝐸6) is
obtained from results of Norton [61]. For 𝑐 = 1/2, our table shows that there are no
99
such finite-dimensional irreducible representations.
Table 3.2: Refined Filtration by Supports for 𝐸6
𝑐 𝑊 ′ 𝜆 ℋ #Irr
1/12 1 triv H𝑞(𝐸6) 24
𝐸6 triv C 1
1/9 1 triv H𝑞(𝐸6) 24
𝐸6 triv C 1
1/8 1 triv H𝑞(𝐸6) 24
𝐷5 triv C 1
1/6 1 triv H𝑞(𝐸6) 20
𝐷4 triv H𝑞2(𝐴2) 2
𝐴5 triv H−1(𝐴1) 1
𝐸6 triv, 𝑉 C 2
1/5 1 triv H𝑞(𝐸6) 23
𝐴4 triv H𝑞(𝐴1) 2
1/4 1 triv H𝑞(𝐸6) 19
𝐴3 triv H−1,𝑞(𝐵2) 3
𝐷4 triv H1(𝐴2) 3
1/3 1 triv H𝑞(𝐸6) 13
𝐴2 triv Z/2Z n H𝑞(𝐴22) 5
𝐴22 triv H1,𝑞(𝐺2) 4
𝐸6 triv, 𝑉,Λ2𝑉 C 3
1/2 1 triv H𝑞(𝐸6) 8
𝐴1 triv H𝑞(𝐴5) 4
𝐴21 triv H1,−1(𝐵3) 4
𝐴31 triv H−1(𝐴1) ⊗ H1(𝐴2) 3
𝐷4 triv, (3, 1) H1(𝐴2) 6
100
Generalized Hecke Algebras for 𝐸7
In this section we produce a table for 𝐸7 analogous to Table 3.2, following the same
conventions. Again, we only list those parameters 𝑐 = 1/𝑑 for positive integers 𝑑 > 1
dividing a fundamental degree of 𝐸7 - the degrees of 𝐸7 are 2, 6, 8, 10, 12, 14, and 18.
The group 𝐸6 has 60 irreducible representations. By inspection, we see that Theorem
3.3.2.14 holds in type 𝐸7.
Note that there are two distinct conjugacy classes of parabolic subgroups of 𝐸7 of
type 𝐴5, while inside 𝐸6 and 𝐸8 there is only 1. We denote a representative of the
conjugacy class of parabolic subgroups of type 𝐴5 appearing already in 𝐸6 by 𝐴′5, and
we denote a representative of the remaining conjugacy class by 𝐴′′5.
There are also two distinct conjugacy classes of parabolic subgroups of 𝐸7 of type
𝐴31, although one of these is absent in Howlett’s table [47]. The two classes can be
distinguished by containment in parabolic subgroups of type 𝐷4; we denote the class
of parabolic subgroups of type 𝐴31 contained in a parabolic subgroup of 𝐷4 by (𝐴3
1)′
and the remaining class by (𝐴31)
′′. The class (𝐴31)
′ is treated in Howlett’s paper; the
complement 𝑁(𝐴31)
′ to (𝐴31)
′ in its normalizer in 𝐸7 acts on h(𝐴31)
′ as a reflection group
of type 𝐵3 × 𝐴1. In the remaining case (𝐴31)
′′, the complement in the normalizer
acts in the fixed space as a reflection group of type 𝐹4. This can be seen as follows.
Computations in GAP3 [36, 58] show that this complement has order 1152. This
group has a decomposition as a semidirect product 𝑁 𝑐𝑜𝑚𝑝 n 𝑁 𝑟𝑒𝑓 , where 𝑁 𝑟𝑒𝑓 is a
real reflection group of rank at most 4 and 𝑁 𝑐𝑜𝑚𝑝 is a finite group acting on 𝑁 𝑟𝑒𝑓
by diagram automorphisms. From the classification of finite reflection groups, the
only possibilities giving rise to groups of order 1152 are 𝑁 𝑟𝑒𝑓 = 𝐹4 and 𝑁 𝑐𝑜𝑚𝑝 = 1,
or 𝑁 𝑟𝑒𝑓 = 𝐷4 and 𝑁 𝑐𝑜𝑚𝑝 = 𝑆3 where 𝑆3 acts on 𝐷4 by the full group of diagram
automorphisms. As (𝐴31)
′′ is contained in parabolic subgroups of different types 𝐴41
and 𝐴1 ×𝐴3 in which it is not self-normalizing, it follows that the hyperplanes in the
reflection representation of 𝑁 𝑟𝑒𝑓 cannot all be conjugate. This rules out 𝑁 𝑟𝑒𝑓 = 𝐷4,
and we conclude that the representation of the complement in the space h(𝐴31)
′′ is
identified with the reflection representation of 𝐹4.
101
When the denominator of 𝑐 is greater than 2, the lowest weights listed for the
finite-dimensional irreducible representations of 𝐸7 were determined by Norton [61],
and we list those representations in the table below. When the denominator of 𝑐 equals
2, we see that there are exactly 7 isomorphism classes of finite-dimensional irreducible
representations of 𝐻𝑐(𝐸7, h). In [45], all but 7 possible lowest weights for these ir-
reducible representations were ruled out. In particular, these 7 representations are
in fact finite-dimensional, and this completes the classification of finite-dimensional
irreducible representations of the rational Cherednik algebras of type 𝐸7.
Table 3.3: Refined Filtration by Supports for 𝐸7
𝑐 𝑊 ′ 𝜆 ℋ #Irr
1/18 1 triv H𝑞(𝐸7) 59
𝐸7 triv C 1
1/14 1 triv H𝑞(𝐸7) 59
𝐸7 triv C 1
1/12 1 triv H𝑞(𝐸7) 58
𝐸6 triv H𝑞3(𝐴1) 2
1/10 1 triv H𝑞(𝐸7) 57
𝐷6 triv H𝑞3(𝐴1) 2
𝐸7 𝑉 C 1
1/9 1 triv H𝑞(𝐸7) 58
𝐸6 triv H1(𝐴1) 2
1/8 1 triv H𝑞(𝐸7) 56
𝐷5 triv H𝑞2(𝐴1) ⊗ H𝑞(𝐴1) 4
1/7 1 triv H𝑞(𝐸7) 58
𝐴6 triv H1(𝐴1) 2
1/6 1 triv H𝑞(𝐸7) 43
𝐷4 triv H𝑞,𝑞2(𝐵3) 6
𝐴5 triv H−1(𝐴21) 1
𝐴′5 triv H−1,𝑞(𝐺2) 3
102
𝐷6 triv H−1(𝐴1) 1
𝐸6 triv, 𝑉 H−1(𝐴1) 2
𝐸7 triv, 𝑉, 𝜙15,7, 𝜙21,6 C 4
1/5 1 triv H𝑞(𝐸7) 54
𝐴4 triv Z/2Z n H𝑞(𝐴2) 6
1/4 1 triv H𝑞(𝐸7) 40
𝐴3 triv H−1,𝑞(𝐵3) ⊗ H𝑞(𝐴1) 10
𝐷4 triv H𝑞,1(𝐵3) 10
1/3 1 triv H𝑞(𝐸7) 32
𝐴2 triv Z/2Z n H𝑞(𝐴5) 14
𝐴22 triv H1,𝑞(𝐺2) ⊗ H1(𝐴1) 8
𝐸6 triv, 𝑉,Λ2𝑉 H1(𝐴1) 6
1/2 1 triv H−1(𝐸7) 12
𝐴1 triv H−1(𝐷6) 6
𝐴21 triv H1,−1(𝐵4) ⊗ H−1(𝐴1) 6
(𝐴31)
′ triv H−1,1(𝐵3) ⊗ H−1(𝐴1) 3
(𝐴31)
′′ triv H1,−1(𝐹4) 9
𝐴41 triv H−1,1(𝐵3) 3
𝐷4 triv, (3, 1) H−1,1(𝐵3) 6
𝐷4 × 𝐴1 triv, (3, 1) ⊗ triv H−1,1(𝐵2) 4
𝐷6 triv, (0, 32), (1, 5), (2, 4) H−1(𝐴1) 4
𝐸7 triv, 𝑉, 𝜙15,7, 𝜙21,6
𝜙27,2, 𝜙35,13, 𝜙189,5 C 7
Generalized Hecke Algebras for 𝐸8
Next we produce a table describing the generalized Hecke algebras arising from 𝐸8.
Again, we only list those parameters 𝑐 = 1/𝑑 for positive integers 𝑑 > 1 dividing
one of the fundamental degrees 2, 8, 12, 14, 18, 20, 24, and 30 of 𝐸8. There are 112
irreducible representations of the group 𝐸8. By inspection, we see that Theorem
103
3.3.2.14 holds in type 𝐸8 as well.
When the denominator of 𝑐 is 2, we see that there are 12 isomorphism classes of
finite-dimensional irreducible representations. Comparing with the lists of “potential”
lowest weights appearing in [45], we are again able to give a complete list of the finite-
dimensional irreducible representations in his case. By similar comparisons with the
results of [45], we are able to obtain such lists in all cases except when the denominator
of 𝑐 is 3, 4 or 18. In those cases, we give the list of “potential” lowest weights from [45]
and the number of those which are in fact finite-dimensional. In each of these three
cases, we see that exactly one of these “potential” finite-dimensional representations is
in fact infinite-dimensional. Furthermore, this problem is resolved fairly easily when
the denominator of 𝑐 is not 3. For denominator 18, Rouquier [63] proved that the
representation 𝐿1/18(𝑉 ) is finite-dimensional, and therefore the remaining “potential”
finite-dimensional representation 𝐿1/18(𝜙28,8), is in fact infinite-dimensional. That
𝐿1/18(𝜙28,8) is infinite-dimensional can be seen independently by the observation that
its lowest eu-weight, 2/3, is not a nonpositive integer. When the denominator of 𝑐 is
4, the entire decomposition matrix for the block of 𝒪1/4(𝐸8, h) containing the simple
object 𝐿1/4(𝜙28,8) can be easily produced following methods of Norton [61, Lemmas
3.5, 3.6], yielding the equality
[𝐿1/4(𝜙28,8)]
= [∆1/4(𝜙28,8)] − [∆1/4(𝜙700,16)] + [∆1/4(𝜙1344,19)] − [∆1/4(𝜙700,28)] + [∆1/4(𝜙28,68)]
in the Grothendieck group of 𝒪1/4(𝐸8, h). It follows that Supp(𝐿1/4(𝜙28,8)) = 𝐸8h𝐷4
and in particular that 𝐿1/4(𝜙28,8) is infinite-dimensional, ruling out this “potential”
finite-dimensional representation.
Table 3.4: Refined Filtration by Supports for 𝐸8
𝑐 𝑊 ′ 𝜆 ℋ #Irr
1/30 1 triv H𝑞(𝐸8) 111
𝐸8 triv C 1
1/24 1 triv H𝑞(𝐸8) 111
104
Table 3.4: continued
𝐸8 triv C 1
1/20 1 triv H𝑞(𝐸8) 111
𝐸8 triv C 1
1/18 1 triv H𝑞(𝐸8) 109
𝐸7 triv H𝑞3(𝐴1) 2
𝐸8 𝑉 C 1
1/15 1 triv H𝑞(𝐸8) 110
𝐸8 triv, 𝑉 C 2
1/14 1 triv H𝑞(𝐸8) 110
𝐸7 triv H𝑞(𝐴1) 2
1/12 1 triv H𝑞(𝐸8) 102
𝐸6 triv H𝑞3,𝑞(𝐺2) 5
𝐷7 triv H−1(𝐴1) 1
𝐸8 triv, 𝜙28,8, 𝜙35,2, 𝜙50,8 C 4
1/10 1 triv H𝑞(𝐸8) 104
𝐷6 triv H𝑞2,𝑞3(𝐵2) 4
𝐸7 𝑉 H−1(𝐴1) 1
𝐸8 triv, 𝑉, 𝜙28,8 C 3
1/9 1 triv H𝑞(𝐸8) 106
𝐸6 triv H1,𝑞(𝐺2) 6
1/8 1 triv H𝑞(𝐸8) 100
𝐷5 triv H𝑞2,𝑞(𝐵3) 9
𝐴7 triv H−1(𝐴1) 1
𝐸8 triv, 𝜙160,7 C 2
1/7 1 triv H𝑞(𝐸8) 108
𝐴6 triv H1,𝑞(𝐴21) 4
1/6 1 triv H𝑞(𝐸8) 75
𝐷4 triv H𝑞2,𝑞(𝐹4) 13
105
Table 3.4: continued
𝐴5 triv H−1,𝑞(𝐺2) ⊗ H−1(𝐴1) 3
𝐷6 triv H1,−1(𝐵2) 2
𝐸6 triv, 𝑉 H−1,𝑞(𝐺2) 6
𝐸7 triv, 𝑉, 𝜙15,2, 𝜙21,6 H−1(𝐴1) 4
𝐸8 triv, 𝑉, 𝜙28,8, 𝜙35,2, 𝜙50,8
𝜙56,19, 𝜙175,12, 𝜙300,8, 𝜙210,4 C 9
1/5 1 triv H𝑞(𝐸8) 96
𝐴4 triv Z/2Z n H𝑞(𝐴4) 12
𝐸8 triv, 𝑉, 𝜙28,8, 𝜙56,19 C 4
1/4 1 triv H𝑞(𝐸8) 69
𝐴3 triv H−1,𝑞(𝐵5) 14
𝐷4 triv H1,𝑞(𝐹4) 20
𝐴23 triv H1,−1(𝐵2) 2
𝐷7 triv, (2, 5) H−1(𝐴1) 2
𝐸8 triv, 𝜙35,2, 𝜙50,8
𝜙210,4, 𝜙350,14 C 5
1/3 1 triv H𝑞(𝐸8) 52
𝐴2 triv Z/2Z n H𝑞(𝐸6) 26
𝐴22 triv Z/2Z n H1,𝑞(𝐺2)
⊗2 14
𝐸6 triv, 𝑉,Λ2𝑉 H1,𝑞(𝐺2) 12
𝐸8 exactly 8 of triv, 𝑉
𝜙28,8, 𝜙35,2, 𝜙50,8, 𝜙160,7
𝜙175,12, 𝜙300,8, 𝜙840,13 C 8
1/2 1 triv H−1(𝐸8) 23
𝐴1 triv H−1(𝐸7) 12
𝐴21 triv H1,−1(𝐵6) 12
𝐴31 triv H−1,1(𝐹4) ⊗ H−1(𝐴1) 9
𝐴41 triv H−1,1(𝐵4) 5
106
Table 3.4: continued
𝐷4 triv, (3, 1) H1,−1(𝐹4) 18
𝐷4 × 𝐴1 triv, (3, 1) ⊗ triv H−1,1(𝐵3) 6
𝐷6 triv, (0, 32), (1, 5), (2, 4) H−1,1(𝐵2) 8
𝐸7 triv, 𝑉, 𝜙15,7, 𝜙21,6
𝜙27,2, 𝜙35,13, 𝜙189,5 H−1(𝐴1) 7
𝐸8 triv, 𝑉, 𝜙28,8, 𝜙35,2,
𝜙50,8, 𝜙175,12, 𝜙300,8, 𝜙210,4,
𝜙560,5, 𝜙840,14, 𝜙1050,10, 𝜙1400,8 C 12
3.3.8 Type 𝐻
Generalized Hecke Algebras for 𝐻3
Next we produce a table describing the generalized Hecke algebras arising from 𝐻3.
Again, we only list those parameters 𝑐 = 1/𝑑 for positive integers 𝑑 > 1 dividing one
of the fundamental degrees 2, 6 and 10 of 𝐻3. There are 10 irreducible representations
of the group 𝐻3. By inspection, we see that Theorem 3.3.2.14 holds in type 𝐻3 as
well.
Table 3.5: Refined Filtration by Supports for 𝐻3
𝑐 𝑊 ′ 𝜆 ℋ #Irr
1/10 1 triv H𝑞(𝐻3) 9
𝐻3 triv C 1
1/6 1 triv H𝑞(𝐻3) 9
𝐻3 triv C 1
1/5 1 triv H𝑞(𝐻3) 8
𝐼2(5) triv H1(𝐴1) 2
1/3 1 triv H𝑞(𝐻3) 8
𝐴2 triv H1(𝐴1) 2
1/2 1 triv H𝑞(𝐻3) 5
107
Table 3.5: continued
𝐴1 triv H−1(𝐴21) 1
𝐴21 triv H−1(𝐴1) 1
𝐻3 triv, 𝑉, 𝑉 C 3
Generalized Hecke Algebras for 𝐻4
Next we produce a table describing the generalized Hecke algebras arising from 𝐻4.
Again, we only list those parameters 𝑐 = 1/𝑑 for positive integers 𝑑 > 1 dividing
one of the fundamental degrees 2, 12, 20 and 30 of 𝐻4. There are 34 irreducible
representations of the group 𝐻4. By inspection, we see that Theorem 3.3.2.14 holds
in type 𝐻4 as well.
When the denominator of 𝑐 is not equal to 2, the list of lowest weights 𝜆 giving
the finite-dimensional irreducible representations of 𝐻𝑐(𝐻4, h) are obtained from re-
sults of Norton [61]. When the denominator is 2, our count below shows that there
are exactly 6 isomorphism classes of finite-dimensional irreducible representations of
𝐻𝑐(𝐻4, h). In [45, Section 5.6], all except 6 of the possible lowest weights 𝜆 for the
finite-dimensional irreducible representations 𝐿𝑐(𝜆) are ruled out, and in particular
those “potential” lowest weights are in fact precisely the lowest weights of the finite-
dimensional 𝐿𝑐(𝜆). This confirms a conjecture of Norton [61] about the classification
of these representations.
Table 3.6: Refined Filtration by Supports for 𝐻4
𝑐 𝑊 ′ 𝜆 ℋ #Irr
1/30 1 triv H𝑞(𝐻4) 33
𝐻4 triv C 1
1/20 1 triv H𝑞(𝐻4) 33
𝐻4 triv C 1
1/15 1 triv H𝑞(𝐻4) 32
𝐻4 triv, 𝑉 C 2
108
Table 3.6: continued
1/12 1 triv H𝑞(𝐻4) 33
𝐻4 triv C 1
1/10 1 triv H𝑞(𝐻4) 29
𝐻3 triv H−1(𝐴1) 1
𝐻4 triv, 𝑉, 𝑉 , 𝜙9,6 C 4
1/6 1 triv H𝑞(𝐻4) 30
𝐻3 triv H−1(𝐴1) 1
𝐻4 triv, 𝑉, 𝑉 C 3
1/5 1 triv H𝑞(𝐻4) 24
𝐼2(5) triv H1,𝑞(𝐼2(10)) 6
𝐻4 triv, 𝑉, 𝜙9,2, 𝜙16,6 C 4
1/4 1 triv H𝑞(𝐻4) 30
𝐴3 triv H−1(𝐴1) 1
𝐻4 triv, 𝜙9,2, 𝜙9,6 C 3
1/3 1 triv H𝑞(𝐻4) 26
𝐴2 triv H1,𝑞(𝐺2) 4
𝐻4 triv, 𝑉, 𝑉 , 𝜙16,3 C 4
1/2 1 triv H−1(𝐻4) 18
𝐴1 triv H−1(𝐻3) 5
𝐴21 triv H1,−1(𝐵2) 2
𝐻3 triv, 𝑉, 𝑉 H−1(𝐴1) 3
𝐻4 triv, 𝑉, 𝑉 , 𝜙9,2, 𝜙9,6, 𝜙25,4 C 6
3.3.9 Type 𝐹4 with Unequal Parameters
In this section we will both prove Theorem 3.3.2.14 for rational Cherednik algebras
of type 𝐹4 and count the number of irreducible representations in 𝒪𝑐1,𝑐2(𝐹4, h) of
each possible support for all values of the parameters 𝑐1, 𝑐2, including in the case of
unequal parameters. As a corollary, comparing with the results of [45], we classify
109
the irreducible finite-dimensional representations of 𝐻1/2,1/2(𝐹4, h). This confirms a
conjecture of Norton [61] about these representations and completes the classification
of the irreducible finite-dimensional representations of rational Cherednik algebras of
type 𝐹4 with equal parameters, the other equal parameter cases having been treated
by Norton. We expect that comparing our counts with the results of [45] completes
the classification of the finite-dimensional irreducible representations for many other
parameter values as well, although we do not perform this comparison here.
First, we fix some notation for the Coxeter group 𝐹4. Our convention will be
that the simple roots of 𝐹4 are labeled 𝑠1, 𝑠2, 𝑠3, 𝑠4 where 𝑠1, 𝑠2 are short simple roots
(with parameter 𝑐1 ∈ C) and 𝑠3, 𝑠4 are long simple roots (with parameter 𝑐2 ∈ C),
and that 𝑠2, 𝑠3 are adjacent in the Dynkin diagram. The 𝐾𝑍 parameters are given
by 𝑝 := 𝑒−2𝜋𝑖𝑐1 and 𝑞 := 𝑒−2𝜋𝑖𝑐2 . We label standard parabolic subgroups, one from
each conjugacy class, as follows:
𝐴′1 := ⟨𝑠1⟩
𝐴1 := ⟨𝑠4⟩
𝐴′1 × 𝐴1 := ⟨𝑠1, 𝑠4⟩
𝐴′2 := ⟨𝑠1, 𝑠2⟩
𝐴2 := ⟨𝑠3, 𝑠4⟩
𝐵2 := ⟨𝑠2, 𝑠3⟩
𝐶3 := ⟨𝑠1, 𝑠2, 𝑠3⟩
𝐵3 := ⟨𝑠2, 𝑠3, 𝑠4⟩
𝐴′1 × 𝐴2 := ⟨𝑠1, 𝑠3, 𝑠4⟩
𝐴′2 × 𝐴1 := ⟨𝑠1, 𝑠2, 𝑠4⟩
Theorem 3.3.2.14 follows in the case 𝑊 = 𝐹4 by application of Remark 3.2.2.3
to the cases in which 𝑊 ′ ⊂ 𝑊 is one of the standard parabolic subgroups appearing
above. In particular, when 𝑊 ′ is a product of Coxeter groups of type 𝐴, the only
irreducible representations 𝜆 of 𝑊 ′ which can appear as the lowest weights of a finite-
dimensional representation 𝐿𝑐(𝜆) are linear characters, i.e. tensor products of either
trivial or sign representations. These representations always extend to representations
of 𝑁𝐹4(𝑊′), and then the statement Theorem 3.3.2.14 follows from Remark 3.2.2.3.
110
The remaining parabolic subgroups 𝑊 ′ are 𝐵2, 𝐵3, and 𝐶3. In each of these cases,
Howlett [47] has shown that the normalizer 𝑁𝐹4(𝑊′) splits as a direct product 𝑊 ′ ×
𝑊 ′′, and Theorem 3.3.2.14 follows by Remark 3.2.2.3 in this case as well.
The classification of the irreducible finite-dimensional representations of the ra-
tional Cherednik algebras 𝐻𝑐(𝑊′, h𝑊 ′) where 𝑊 ′ ⊂ 𝐹4 is one of the proper nontrivial
parabolic subgroups appearing above is well known. Applying our results to each of
these cases, we can count the number of irreducible representations in 𝒪𝑐(𝐹4, h) with
support labeled by any of these parabolic subgroups. This information is presented
in the following tables; the left column of each table specifies certain conditions on
the parameters 𝑐1, 𝑐2 in terms of the 𝐾𝑍 parameters 𝑝 = 𝑒−2𝜋𝑖𝑐1 and 𝑞 = 𝑒−2𝜋𝑖𝑐2 ,
and the right column of each table gives the number of irreducible representations in
𝒪𝑐1,𝑐2(𝐹4, h) with support labeled by the parabolic subgroup appearing in the title of
the table. Here Φ𝑑 denotes the 𝑑𝑡ℎ cyclotomic polynomial.
Table 3.7: Simple Modules for 𝐻𝑐1,𝑐2(𝐹4) labeled by 𝐴′1
condition # simples labeled
𝑝 = −1 and Φ1(𝑞)Φ2(𝑞)Φ3(𝑞)Φ4(𝑞) = 0 9
𝑝 = −1 and Φ1(𝑞)Φ2(𝑞) = 0 3
𝑝 = −1 and Φ3(𝑞) = 0 6
𝑝 = −1 and Φ4(𝑞) = 0 8
otherwise 0
Table 3.8: Simple Modules for 𝐻𝑐1,𝑐2(𝐹4) labeled by 𝐴1
condition # simples labeled
𝑞 = −1 and Φ1(𝑝)Φ2(𝑝)Φ3(𝑝)Φ4(𝑝) = 0 9
𝑞 = −1 and Φ1(𝑝)Φ2(𝑝) = 0 3
𝑞 = −1 and Φ3(𝑝) = 0 6
𝑞 = −1 and Φ4(𝑝) = 0 8
otherwise 0
111
Table 3.9: Simple Modules labeled by 𝐴′1 × 𝐴1
condition # simples labeled
𝑝 = 𝑞 = −1 1
otherwise 0
Table 3.10: Simple Modules labeled by 𝐴′2
condition # simples labeled
Φ3(𝑝) = 0 and Φ2(𝑞)Φ3(𝑞)Φ6(𝑞)Φ12(𝑞) = 0 6
Φ3(𝑝) = 0 and Φ2(𝑞)Φ6(𝑞) = 0 3
Φ3(𝑝) = 0 and Φ3(𝑞) = 0 4
Φ3(𝑝) = 0 and Φ12(𝑞) = 0 5
otherwise 0
Table 3.11: Simple Modules labeled by 𝐴2
condition # simples labeled
Φ3(𝑞) = 0 and Φ2(𝑝)Φ3(𝑝)Φ6(𝑝)Φ12(𝑝) = 0 6
Φ3(𝑞) = 0 and Φ2(𝑝)Φ6(𝑝) = 0 3
Φ3(𝑞) = 0 and Φ3(𝑝) = 0 4
Φ3(𝑞) = 0 and Φ12(𝑝) = 0 5
otherwise 0
Table 3.12: Simple Modules labeled by 𝐵2
condition # simples labeled
𝑝 = −1 and 𝑞 = −1 2
𝑝±1 = −𝑞 and Φ3(𝑞)Φ6(𝑞) = 0 2
𝑝±1 = −𝑞 and Φ3(𝑞)Φ6(𝑞) = 0 4
112
Table 3.12: continued
otherwise 0
Table 3.13: Simple Modules labeled by 𝐵3
condition # simples labeled
𝑝 = −𝑞±2 and Φ1(𝑞)Φ6(𝑞)Φ12(𝑞) = 0 2
𝑝 = −𝑞±2 and Φ6(𝑞)Φ12(𝑞) = 0 1
𝑝 = −1 = −𝑞±2 and 𝑞 = 1 3
𝑝 = −1 and Φ3(𝑞) = 0 1
otherwise 0
Table 3.14: Simple Modules labeled by 𝐶3
condition # simples labeled
𝑞 = −𝑝±2 and Φ1(𝑝)Φ6(𝑝)Φ12(𝑝) = 0 2
𝑞 = −𝑝±2 and Φ6(𝑝)Φ12(𝑝) = 0 1
𝑞 = −1 = −𝑝±2 and 𝑝 = 1 3
𝑞 = −1 and Φ3(𝑝) = 0 1
otherwise 0
Table 3.15: Simple Modules labeled by 𝐴′1 × 𝐴2
condition # simples labeled
𝑝 = −1 and Φ3(𝑞) = 0 1
otherwise 0
113
Table 3.16: Simple Modules labeled by 𝐴′2 × 𝐴1
condition # simples labeled
Φ3(𝑝) = 0 and 𝑞 = −1 1
otherwise 0
To count the irreducible finite-dimensional representations for given parameters
(𝑐1, 𝑐2), we need first to count the irreducible representations in 𝒪𝑐1,𝑐2(𝐹4, h) of full
support, which is achieved by counting the number of irreducible representations
of the Hecke algebra H𝑝,𝑞(𝐹4). The following table, produced by computations in
CHEVIE in GAP3 [36, 58] and in Mathematica, gives this number of irreducible
representations in all cases up to obvious symmetries. In particular, the isomorphism
class as a C-algebra of the Hecke algebra H𝑝,𝑞(𝐹4) does not change when the order
of the parameters 𝑝, 𝑞 is reversed or when either of the parameters is replaced with
its inverse. The following table then gives the number of irreducible representations
of H𝑝,𝑞(𝐹4) for a complete set of parameters 𝑝, 𝑞 ∈ C× up to these symmetries. The
first column specifies a hypersurface in the 𝑝, 𝑞-plane at which the Hecke algebra
H𝑝,𝑞(𝐹4) is not semisimple, and the second column specifies additional conditions on
𝑞, and the final column gives the corresponding number of irreducible representations
of H𝑝,𝑞(𝐹4).
Table 3.17: Irreducible Representations of H𝑝,𝑞(𝐹4)
condition 𝑞 condition # simples
𝑝 = −1 𝑞 = 1 9
𝑞 = −1 8
𝑞 = roots of unity order 3 11
𝑞 = roots of unity of order 4 or 6 14
otherwise 15
𝑝±1 + 𝑞 = 0 𝑞 = ±1 9
𝑞 = roots of unity of order 3 or 6 13
𝑞 = roots of unity of order 8 18
114
Table 3.17: continued
otherwise 19
𝑝±1 + 𝑞2 = 0 𝑞 = 1 9
𝑞 = −1 8
𝑞 = roots of unity of order 3 13
𝑞 = roots of unity of order 4 or 6 20
𝑞 = roots of unity of order 9 22
𝑞 = roots of unity of order 10 21
𝑞 = roots of unity of order 12 17
otherwise 23
𝑝±1 = 𝑖𝑞 𝑞 = 1 or 𝑞 = −𝑖 20
𝑞 = −1 14
𝑞 = roots of unity of order 3 17
𝑞 = 𝑖 14
𝑞 = roots of unity of order 6 23
𝑞 ∈ 𝑒2𝜋𝑖/8, 𝑒5·2𝜋𝑖/8 18
𝑞 ∈ 𝑒2𝜋𝑖/12, 𝑒5·2𝜋𝑖/12 17
𝑞 ∈ 𝑒7·2𝜋𝑖/12, 𝑒11·2𝜋𝑖/12 23
𝑞 ∈ 𝑒7·2𝜋𝑖/24, 𝑒11·2𝜋𝑖/24, 𝑒19·2𝜋𝑖/24, 𝑒23·2𝜋𝑖/24 23
otherwise 24
Φ3(𝑝) = 0 𝑞 = −1 11
𝑞 = roots of unity of order 3 15
𝑞 = roots of unity of order 6 13
𝑞 = roots of unity of order 12 17
otherwise 19
Φ6(𝑝) = 0 𝑞 = 1 22
𝑞 = −1 14
𝑞 = roots of unity of order 3 13
𝑞 = roots of unity of order 6 20
115
Table 3.17: continued
𝑞 = roots of unity of order 12 23
otherwise 24
𝑝±1 = 𝑒2𝜋𝑖/6𝑞 𝑞 = 1 22
𝑞 = −1 11
𝑞 = 𝑒2𝜋𝑖/3 11
𝑞 = 𝑒2·2𝜋𝑖/3 13
𝑞 = 𝑒2𝜋𝑖/6 13
𝑞 = 𝑒5·2𝜋𝑖/6 22
𝑞 ∈ 𝑒2𝜋𝑖/9, 𝑒4·2𝜋𝑖/9, 𝑒7·2𝜋𝑖/9 22
𝑞 ∈ 𝑒2𝜋𝑖/18, 𝑒7·2𝜋𝑖/18, 𝑒13·2𝜋𝑖/18 22
𝑞 ∈ 𝑒2𝜋𝑖/24, 𝑒7·2𝜋𝑖/24, 𝑒13·2𝜋𝑖/24, 𝑒19·2𝜋𝑖/24 23
otherwise 24
otherwise 25
The following table consolidates the information in the previous tables and counts
the number of irreducible finite-dimensional representations of 𝐻𝑐1,𝑐2(𝐹4, h) for all pa-
rameters 𝑐1, 𝑐2 in terms of the 𝐾𝑍 parameters 𝑝 and 𝑞, up to the same symmetries
mentioned above, i.e. up to exchanging and inverting 𝑝 and 𝑞, for which the category
𝒪𝑐1,𝑐2(𝐹4, h) is not semisimple. The first column specifies conditions on the param-
eters 𝑝 and 𝑞. The remaining columns give the counts of the number of irreducible
representations in 𝒪𝑐1,𝑐2(𝐹4, h), for any parameters 𝑐1, 𝑐2 satisfying 𝑝 = 𝑒−2𝜋𝑖𝑐1 and
𝑞 = 𝑒−2𝜋𝑖𝑐2 , with support labeled by the parabolic subgroup appearing at the top of
the column. In particular, the last column, labeled by 𝐹4, counts finite-dimensional
representations. This last column is obtained from the others by subtracting their
sum from 25, the number of irreducible complex representations of the Coxeter group
𝐹4. For those conditions on 𝑝 and 𝑞 specifying a certain finite set of points, we give
a defining ideal for these points; for example, (Φ2(𝑝),Φ3(𝑞)) specifies that 𝑝 = −1
and that 𝑞 is a primitive cube root of unity. The remaining conditions specify certain
curves with certain exceptional points removed.
116
Tabl
e3.
18:
Mod
ules
ofG
iven
Supp
ort
in𝐻𝑐 1,𝑐
2(𝐹
4)
cond
itio
n1
𝐴′ 1
𝐴1
𝐴′ 1×𝐴
1𝐴
′ 2𝐴
2𝐵
2𝐵
3𝐶
3𝐴
′ 1×𝐴
2𝐴
′ 2×𝐴
1𝐹4
(Φ1(𝑝
),Φ
2(𝑞
))9
.3
..
.4
.3
..
6
(Φ1(𝑝
),Φ
4(𝑞
))20
..
..
..
2.
..
3
(Φ1(𝑝
),Φ
6(𝑞
))22
..
..
..
..
..
3
(Φ2(𝑝
),Φ
2(𝑞
))8
33
1.
.2
22
..
4
(Φ2(𝑝
),Φ
3(𝑞
))11
6.
..
3.
1.
1.
3
(Φ2(𝑝
),Φ
4(𝑞
))14
8.
..
..
..
..
3
(Φ2(𝑝
),Φ
6(𝑞
))14
9.
..
..
..
..
2
(Φ3(𝑝
),Φ
3(𝑞
))15
..
.4
4.
..
..
2
(Φ3(𝑝
),Φ
6(𝑞
))13
..
.3
.2
.2
..
5
(Φ3(𝑝
),Φ
12(𝑞
))17
..
.5
..
1.
..
2
(Φ4(𝑝
),Φ
4(𝑞
))19
..
..
.4
..
..
2
(Φ6(𝑝
),Φ
6(𝑞
))20
..
..
..
11
..
3
(Φ6(𝑝
),Φ
12(𝑞
))23
..
..
..
..
..
2
(Φ8(𝑝
),Φ
8(𝑞
),
Φ1(𝑝𝑞)
Φ1(𝑝𝑞−
1))
24.
..
..
..
..
.1
(Φ8(𝑝
),Φ
8(𝑞
),
Φ2(𝑝𝑞)
Φ2(𝑝𝑞−
1))
18.
..
..
4.
..
.3
117
Tabl
e3.
18:
cont
inue
d
(Φ9(𝑝
),Φ
18(𝑞
),
Φ2(𝑝
2𝑞)
Φ2(𝑝
2𝑞−
1))
22.
..
..
..
2.
.1
(Φ10(𝑝
),Φ
10(𝑞
),
Φ2(𝑝𝑞2
)Φ2(𝑝
−1𝑞2
))21
..
..
..
22
..
.
(Φ12(𝑝
),Φ
12(𝑞
),
Φ1(𝑝𝑞)
Φ1(𝑝
−1𝑞)
)24
..
..
..
..
..
1
(Φ24(𝑝
),Φ
24(𝑞
),
Φ4(𝑝𝑞),Φ
6(𝑝
−1𝑞)
)23
..
..
..
..
..
2
Φ2(𝑝
)=
0an
d
(𝑞6−
1)Φ
4(𝑞
)=
015
9.
..
..
..
..
1
Φ3(𝑝
)=
0an
d
Φ2(𝑞
)Φ3(𝑞
)Φ6(𝑞
)·
Φ12(𝑞
)=
019
..
.6
..
..
..
.
Φ6(𝑝
)=
0an
d
(𝑞6−
1)Φ
12(𝑞
)=
024
..
..
..
..
..
1
Φ2(𝑝𝑞)
=0
and
(𝑞6−
1)Φ
8(𝑞
)=
019
..
..
.4
..
..
2
Φ2(𝑝𝑞2
)=
0an
d
118
Tabl
e3.
18:
cont
inue
d
(𝑞12−
1)Φ
9(𝑞
)·
Φ10(𝑞
)=
023
..
..
..
2.
..
.
Φ4(𝑝𝑞)
=0
and
(𝑞12−
1)Φ
2(𝑝𝑞−
1)·
Φ6(𝑝𝑞−
1)=
024
..
..
..
..
..
1
Φ6(𝑝𝑞)
=0
and
(𝑞6−
1)Φ
4(𝑝𝑞−
1)·
Φ2(𝑝𝑞−
2)·
Φ2(𝑝
−2𝑞)
=0
24.
..
..
..
..
.1
119
Table 3.18 shows in particular that there are exactly 4 distinct irreducible finite-
dimensional representations of the rational Cherednik algebra 𝐻1/2,1/2(𝐹4, h). In no-
tation compatible with the labeling of the irreducible representations of the Coxeter
group 𝐹4 in GAP3, it is shown in [45, Section 5.7.2] that with equal parameters
𝑐1 = 𝑐2 = 1/2 the lowest weights 𝜆 such that the representation 𝐿1/2,1/2(𝜆) is finite-
dimensional are among the representations 𝜙1,0, 𝜙′2,4, 𝜙
′′2,4, and 𝜙9,2. In particular, we
have the following corollary, confirming Norton’s conjecture on the classification of
the irreducible finite-dimensional representations of 𝐻1/2,1/2(𝐹4, h):
Corollary 3.3.9.1. The set of irreducible representations 𝜆 of the Coxeter group 𝐹4
for which the irreducible lowest weight representation 𝐿1/2,1/2(𝐹4, h) is finite dimen-
sional is
𝜙1,0, 𝜙′2,4, 𝜙
′′2,4, 𝜙9,2.
3.3.10 Type 𝐼
The only remaining case in which Theorem 3.3.2.14 needs to be verified is the case
of the irreducible Coxeter groups of type 𝐼, i.e. the dihedral groups. For 𝑑 ≥ 3,
let 𝐼2(𝑑) denote the dihedral Coxeter group with 2𝑑 elements. Chmutova [13] has
computed character formulas for all irreducible representations 𝐿𝑐(𝜆) in the category
𝒪𝑐(𝐼2(𝑑), h) for all parameters 𝑐, and in particular the supports of all irreducible
representations in 𝒪𝑐(𝐼2(𝑑), h) are known in this case, so we will not produce tables
counting the number of irreducible representations of given support in this case.
To see that Theorem 3.3.2.14 holds in type 𝐼, we need only consider the case of
rank 1 parabolic subgroups, i.e. those of type 𝐴1. Let 𝑑 ≥ 3 be an integer and let
𝑊 ′ ⊂ 𝐼2(𝑑) be a parabolic subgroup of type 𝐴1. If the parameter 𝑐1 attached to
the reflection generating 𝐴1 does not satisfy 𝑐1 ∈ 1/2 + Z, the rational Cherednik
algebra 𝐻𝑐1(𝑊′, h𝑊 ′) does not admit any nonzero finite-dimensional representations.
Otherwise, there is a unique finite-dimensional representation 𝐿 of 𝐻𝑐1(𝑊′, h𝑊 ′) with
lowest weight triv or sgn, depending on whether 𝑐1 > 0 or 𝑐1 < 0, respectively. In
either case, 𝐿 is 𝑁𝐼2(𝑑)(𝑊′)-equivariant by Remark 3.2.2.3. This completes the proof
120
of Theorem 3.3.2.14.
121
122
Chapter 4
The Dunkl Weight Function
4.1 Introduction
In addition to characters, another interesting class of invariants attached to irreducible
representations of rational Cherednik algebras are the signature characters. When the
parameter 𝑐 is real, i.e. 𝑐(𝑠) = 𝑐(𝑠−1) for all reflections 𝑠 ∈ 𝑆, each standard module
∆𝑐(𝜆) admits a 𝑊 -invariant, graded, contravariant Hermitian form 𝛽𝑐,𝜆, unique up
to scaling by a positive real number. The kernel of the form 𝛽𝑐,𝜆 coincides with
the unique maximal proper submodule of ∆𝑐(𝜆), and in particular 𝛽𝑐,𝜆 descends to
the irreducible quotient 𝐿𝑐(𝜆). Similarly to the definition of the character of 𝐿𝑐(𝜆),
the signature character of 𝐿𝑐(𝜆) is the generating function recording the signature
of the form 𝛽𝑐,𝜆 in each finite-dimensional graded component of 𝐿𝑐(𝜆). Determining
explicit formulas for these signature characters and describing the set of parameters
𝑐 such that 𝛽𝑐,𝜆 is unitary has proved very difficult, with complete results available
in only a limited collection of cases, e.g. [33] and [70] for the type 𝐴 case and [44]
for the classical and cyclotomic types. The study of the forms 𝛽𝑐,𝜆 can be viewed
as an analogue for rational Cherednik algebras of the study of unitarizability (or,
more generally, of signatures of invariant Hermitian forms) of representations of real
reductive groups considered in detail by Adams, van Leeuwen, Trapa, and Vogan [1].
When 𝑊 is a finite real reflection group, the rational Cherednik algebra con-
tains a natural sl2-triple e, f ,h ∈ 𝐻𝑐(𝑊, h). The element f acts by a degree -2
123
operator, and hence nilpotently, on any representation 𝑀 ∈ 𝒪𝑐(𝑊, h). In particu-
lar, its exponential exp(f) is well-defined and preserves the natural filtration on any
𝑀 ∈ 𝒪𝑐(𝑊, h). In particular, the determination of the signatures of the Gaussian in-
ner product 𝛾𝑐,𝜆(𝑣, 𝑣′) := 𝛽𝑐,𝜆(exp(f)𝑣, exp(f)𝑣′), considered in [10] and [33, Definition
4.5], in the filtered pieces of ∆𝑐(𝜆) is equivalent to the determination of the signature
character of 𝐿𝑐(𝜆). As it happens, the Gaussian inner product 𝛾𝑐,𝜆 appears easier to
study than the contravariant form 𝛽𝑐,𝜆 itself.
The main purpose of this chapter, generalizing previous results of Dunkl [19, 20,
21], is to introduce a fundamental object, the Dunkl weight function 𝐾𝑐,𝜆, for the
study of the Gaussian inner product 𝛾𝑐,𝜆. In fact, it is natural to consider a slightly
more general context in which the parameter 𝑐 is not required to be real; for general
𝑐, 𝛾𝑐,𝜆 is a sesquilinear pairing
𝛾𝑐,𝜆 : ∆𝑐(𝜆) × ∆𝑐†(𝜆) → C,
where 𝑐† is the parameter given by 𝑐†(𝑠) = 𝑐(𝑠−1). By the PBW theorem for rational
Cherednik algebras [29, Theorem 1.3], the standard module ∆𝑐(𝜆) can be identified
with C[h] ⊗ 𝜆 as a C[h] o C𝑊 -module, independently of 𝑐, where C[h] denotes the
algebra of polynomial functions on the reflection representation h. With respect to
this identification, and having chosen a 𝑊 -invariant positive-definite Hermitian form
(𝑣1, 𝑣2) ↦→ 𝑣†2𝑣1 on 𝜆, we have the following result, where hR denotes the real reflection
representation of 𝑊 :
Theorem. For any finite Coxeter group 𝑊 and irreducible representation 𝜆 of 𝑊 ,
there is a unique family 𝐾𝑐,𝜆, holomorphic in 𝑐 ∈ p, of EndC(𝜆)-valued tempered
distributions on hR such that the following integral representation of the Gaussian
pairing 𝛾𝑐,𝜆 holds for all 𝑐 ∈ p:
𝛾𝑐,𝜆(𝑃,𝑄) =
∫hR
𝑄(𝑥)†𝐾𝑐,𝜆(𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥 for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆.
This theorem, along with additional properties of 𝐾𝑐,𝜆, appears in Theorem 4.3.3.1
124
and is proved in Sections 4.3.4-4.3.6. The family of tempered distributions 𝐾𝑐,𝜆
appearing in the theorem above will be called the Dunkl weight function. The case in
which 𝜆 is the trivial representation, including the extension of the weight function
to arbitrary 𝑐 as a tempered distribution, has been studied previously in detail by
Etingof [28], leading to the complete determination of the parameters 𝑐 for which
𝐿𝑐(triv) is finite-dimensional for Coxeter groups 𝑊 . For dihedral groups 𝑊 , Dunkl
has provided explicit formulas for the matrix entries of 𝐾𝑐,𝜆 for 𝑐 small and real.
Additionally, for real parameters 𝑐 = 𝑐†, the Dunkl weight function 𝐾𝑐,𝜆 pro-
vides a bridge between the study of invariant Hermitian forms on representations of
rational Cherednik algebras and of associated finite Hecke algebras via the Knizhnik-
Zamolodchikov (KZ) functor. The KZ functor, introduced by Ginzburg, Guay, Op-
dam, and Rouquier [38], is an exact functor 𝐾𝑍 : 𝒪𝑐(𝑊, h) → H𝑞(𝑊 )-mod𝑓.𝑑. from
the category 𝒪𝑐(𝑊, h) to the category H𝑞(𝑊 )-mod𝑓.𝑑. of finite-dimensional represen-
tations of the Hecke algebra H𝑞(𝑊 ) attached to the Coxeter group 𝑊 at a particular
parameter 𝑞. The construction of the KZ functor depends on a choice of point 𝑥0 in
the complement h𝑟𝑒𝑔 of the reflection hyperplanes in h, with any two points giving
rise to isomorphic functors. Let 𝐾𝑍𝑥0 denote the functor associated to the point
𝑥0 ∈ h𝑟𝑒𝑔. As a vector space, 𝐾𝑍𝑥0(∆𝑐(𝜆)) is naturally identified with the vector
space 𝜆 itself. In Theorem 4.3.3.1 we will see that the restriction of the distribu-
tion 𝐾𝑐,𝜆 to hR,𝑟𝑒𝑔 := hR ∩ h𝑟𝑒𝑔 is given by integration against an analytic function
with values in EndC(𝜆), and, with respect to the identification 𝐾𝑍𝑥0(∆𝑐(𝜆)) ∼=C 𝜆,
the value 𝐾𝑐,𝜆(𝑥0) at any 𝑥0 ∈ hR,𝑟𝑒𝑔 determines a braid group invariant Hermitian
form on 𝐾𝑍𝑥0(∆𝑐(𝜆)). In Section 4.4, we exploit this relationship between invariant
Hermitian forms on ∆𝑐(𝜆) and 𝐾𝑍𝑥0(∆𝑐(𝜆)) to show that the KZ functor preserves
signatures, and hence unitarizability, in an appropriate sense made precise in Theorem
4.4.0.1.
This chapter is organized as follows. In Section 4.2, we recall the definition of
signature characters, introduce Janzten filtrations on standard modules ∆𝑐(𝜆), and
use the Jantzen filtrations to prove that the shifted signature characters are rational
functions, allowing for the definition of the asymptotic signature 𝑎𝑐,𝜆 ∈ Q ∩ [−1, 1]
125
recording the limiting behavior of the signatures of 𝛽𝑐,𝜆 in graded components of high
degree. In Section 4.3, we state and prove the main theorem on the existence of the
Dunkl weight function. Finally, in Section 4.4, we use the Dunkl weight function and
its properties established in Section 4.3 to prove Theorem 4.4.0.1 on the compatibility
of the KZ functor with signatures. In Section 4.5, we discuss related conjectures and
further directions.
4.2 Signature Characters and the Jantzen Filtration
4.2.1 Jantzen Filtrations on Standard Modules
Let 𝑊 be a finite complex reflection group with reflection representation h. Let
𝑆 ⊂ 𝑊 be the set of complex reflections in 𝑊 , and let p be the C-vector space
of 𝑊 -invariant functions 𝑐 : 𝑆 → C. For any 𝑐 ∈ p, let 𝑐† ∈ p be defined by
𝑐†(𝑠) = 𝑐(𝑠−1). Refer to any 𝑐 ∈ p satisfying 𝑐 = 𝑐† as real, and let pR be the R-vector
space pR := 𝑐 ∈ p : 𝑐 = 𝑐†.
Recall that p is the parameter space for the family of rational Cherednik algebras
attached to (𝑊, h); we denote the rational Cherednik algebra attached to (𝑊, h) and
parameter 𝑐 ∈ p by 𝐻𝑐(𝑊, h). For each irreducible representation 𝜆 ∈ Irr(𝑊 ), we
have the associated standard module
∆𝑐(𝜆) := 𝐻𝑐(𝑊, h) ⊗C𝑊nC[h*] 𝜆
over 𝐻𝑐(𝑊, h). As a Z≥0-graded C-vector space, ∆𝑐(𝜆) is naturally identified with
C[h] ⊗ 𝜆 independently of 𝑐.
Fix a 𝑊 -invariant positive definite Hermitian form (·, ·)𝜆 on 𝜆 - we will take the
convention that Hermitian forms are conjugate-linear in the second factor. Such a
form is uniquely determined up to R>0-scaling. Similarly, fix a 𝑊 -invariant positive-
definite Hermitian form (·, ·)h on h. This determines the conjugate-linear isomorphism
𝑇 : h → h* given by
𝑇 (𝑦)(𝑥) := (𝑥, 𝑦)h.
126
When the parameter 𝑐 ∈ p is real, i.e. 𝑐 = 𝑐†, the standard module ∆𝑐(𝜆) admits
a unique 𝑊 -invariant Hermitian form 𝛽𝑐,𝜆 [33, Proposition 2.2] such that the con-
travariance condition
𝛽𝑐,𝜆(𝑦𝑣, 𝑣′) = 𝛽𝑐,𝜆(𝑣, 𝑇 (𝑦)𝑣)
holds for all 𝑣, 𝑣′ ∈ ∆𝑐(𝜆) and 𝑦 ∈ h and that coincides with (·, ·)𝜆 in degree 0.
As above, regarding the standard modules ∆𝑐(𝜆) for various 𝑐 as the same Z≥0-
graded vector space ∆(𝜆), we may view the forms 𝛽𝑐,𝜆 on ∆(𝜆) as an algebraic family
of Hermitian forms 𝛽𝑐,𝜆[𝑑], parameterized by 𝑐 ∈ pR in a polynomial manner, on each
finite-dimensional graded component ∆(𝜆)[𝑑]. In particular, we are naturally led to
consider Jantzen filtrations, as follows.
Let 𝑐0, 𝑐1 ∈ pR be real parameters such that there exists 𝛿 > 0 such that 𝛽𝑐(𝑡),𝜆
is nondegenerate for all 𝑐(𝑡) := 𝑐0 + 𝑡𝑐1 with 𝑡 ∈ (−𝛿, 𝛿)∖0. The finite-dimensional
C-vector spaces ∆(𝜆)[𝑑] for 𝑛 ≥ 0 along with the polynomial families of Hermitian
forms 𝛽𝑐(𝑡),𝜆[𝑑] satisfy the conditions of [71, Definition 3.1]. In particular, for each
𝑑 ≥ 0 define the finite descending filtration
∆(𝜆)[𝑑] = ∆(𝜆)[𝑑]≥0 ⊃ ∆(𝜆)[𝑑]≥1 ⊃ · · · ⊃ ∆(𝜆)[𝑑]≥𝑁 = 0
on ∆(𝜆)[𝑑] as follows (to simplify the notation, we do not include the choice of 𝑐0, 𝑐1
in the notation for the filtration, but of course this filtration and its properties may
be dependent on this choice). Let ∆(𝜆)[𝑑]≥𝑘 consist of those vectors 𝑣 ∈ ∆(𝜆)[𝑑] such
that there is some 𝜖 > 0 and an analytic function 𝑓𝑣 : (−𝜖, 𝜖) → ∆(𝜆)[𝑑] satisfying
𝑓𝑣(0) = 𝑣 and such that the analytic function
𝑡 ↦→ 𝛽𝑐(𝑡),𝜆[𝑑](𝑓𝑣(𝑡), 𝑣′)
vanishes at least to order 𝑘 for all 𝑣′ ∈ ∆𝑐(𝜆)[𝑑] (clearly, one may equivalently consider
only polynomial functions 𝑓𝑣). For each 𝑘 ≥ 0, define the Hermitian form 𝛽𝑐0,𝜆[𝑑]≥𝑘
127
on ∆(𝜆)[𝑑]≥𝑘 by
𝛽𝑐0,𝜆[𝑑]≥𝑘(𝑣, 𝑣′) = lim𝑡→0
1
𝑡𝑘𝛽𝑐(𝑡),𝜆[𝑑](𝑓𝑣(𝑡), 𝑓𝑣′(𝑡)),
where 𝑣, 𝑣′ ∈ ∆(𝜆)[𝑑]≥𝑘 and 𝑓𝑣, 𝑓𝑣′ are any analytic functions as above (this limit does
not depend on the choice of such 𝑓𝑣, 𝑓𝑣′). We then have the following theorem:
Theorem 4.2.1.1. (Jantzen [50, 5.1], Vogan [71, Theorem 3.2]) The radical of the
Hermitian form 𝛽𝑐0,𝜆[𝑑]≥𝑘 is precisely ∆(𝜆)[𝑑]≥𝑘+1.
In particular, the Hermitian form 𝛽𝑐0,𝜆[𝑑]≥𝑘 descends to a nondegenerate Hermitian
form on each filtration subquotient ∆(𝜆)[𝑑](𝑘) := ∆(𝜆)[𝑑]≥𝑘/∆(𝜆)[𝑑]≥𝑘+1. Denote this
induced nondegenerate Hermitian form by 𝛽𝑐0,𝜆[𝑑](𝑘). For all 𝑘, 𝑑 ≥ 0, let 𝑝(𝑘)𝑑 (resp.
𝑞(𝑘)𝑑 ) be the dimension of a maximal positive definite (resp., negative definite) subspace
of ∆(𝜆)[𝑑](𝑘) with respect to the form 𝛽𝑐0,𝜆[𝑑](𝑘). Note that for any fixed 𝑑 ≥ 0 we
have 𝑝(𝑘)𝑑 = 𝑞(𝑘)𝑑 = 0 for all sufficiently large 𝑘.
We will take the convention that the signature of a Hermitian form 𝛽 on a finite
dimensional C-vector space 𝑉 is the integer 𝑝 − 𝑞, where 𝑝 (respectively, 𝑞) is the
dimension of any maximal positive-definite (respectively, negative-definite) subspace
of 𝑉 with respect to 𝛽. For example, in the context of the previous paragraph,
𝑝(𝑘)𝑑 − 𝑞
(𝑘)𝑑 is the signature of the form 𝛽𝑐0,𝜆[𝑑](𝑘). If the form 𝛽 is non-degenerate then
the dimension of 𝑉 and the quantity 𝑝 − 𝑞 determines the tuple (𝑝, 𝑞) considered to
be the signature of 𝛽 in some references.
Proposition 4.2.1.2. (Vogan [71, Proposition 3.3])
(a) For all small positive 𝑡 (i.e. 𝑡 ∈ (0, 𝛿)) and any 𝑑 ≥ 0, the signature of the
nondegenerate Hermitian form 𝛽𝑐(𝑡),𝜆[𝑑] on ∆(𝜆)[𝑑] is
∑𝑘≥0
𝑝(𝑘)𝑑 −
∑𝑘≥0
𝑞(𝑘)𝑑 .
(b) Similarly, for all small negative 𝑡 (i.e. 𝑡 ∈ (−𝛿, 0)) and any 𝑑 ≥ 0, the signature
128
of 𝛽𝑐(𝑡),𝜆[𝑑] is
(∑𝑘 even
𝑝(𝑘)𝑑 +
∑𝑘 odd
𝑞(𝑘)𝑑
)−
(∑𝑘 odd
𝑝(𝑘)𝑑 +
∑𝑘 even
𝑞(𝑘)𝑑
).
Define the descending filtration
∆(𝜆) = ∆(𝜆)≥0 ⊃ ∆(𝜆)≥1 ⊃ · · ·
by
∆(𝜆)≥𝑘 :=⨁𝑑≥0
∆(𝜆)[𝑑]≥𝑘 ⊂⨁𝑑≥0
∆(𝜆)[𝑑] = ∆(𝜆).
Lemma 4.2.1.3. The filtration of ∆𝑐0(𝜆) by the subspaces ∆(𝜆)≥𝑘 is a filtration by
𝐻𝑐0(𝑊, h)-submodules. We have ∆(𝜆)≥𝑘 = 0 for sufficiently large 𝑘.
Proof. Let 𝑣 ∈ ∆(𝜆)[𝑑]≥𝑘, and let 𝑓𝑣 : (−𝜖, 𝜖) → ∆(𝜆)[𝑑] be as in the definition of the
filtration, exhibiting that 𝑣 ∈ ∆(𝜆)[𝑑]≥𝑘. Then for any homogeneous ℎ ∈ 𝐻𝑐0(𝑊, h)
of degree 𝑑′, viewing ℎ ∈ 𝐻𝑐(𝑡)(𝑊, h) for all 𝑡 ∈ R via the PBW basis, the path ℎ𝑓𝑣
exhibits ℎ𝑣 as an element of ∆(𝜆)[𝑑 + 𝑑′]≥𝑘. In particular, ∆(𝜆)≥𝑘 is a 𝐻𝑐0(𝑊, h)-
submodule of ∆(𝜆). By the finite-length property of 𝐻𝑐0(𝑊, h)-modules in category
𝒪𝑐0(𝑊, h), it follows that the filtration ∆(𝜆)≥𝑘 stabilizes in 𝑘. For any fixed 𝑑 we
have ∆(𝜆)[𝑑]≥𝑘 = 0 for any 𝑘 sufficiently large, and it follows that ∆(𝜆)≥𝑘 = 0 for all
sufficiently large 𝑘.
We refer to the filtration of ∆𝑐0(𝜆) appearing in Lemma 4.2.1.3 as the Jantzen
filtration of ∆𝑐0(𝜆). Note that this Jantzen filtration depends on the choice of the
additional parameter 𝑐1 ∈ pR determining the direction for the deformation.
4.2.2 Hermitian Duals
In this section we will introduce Hermitian duals in the setting of rational Cherednik
algebras, analogous to the Hermitian duals considered by Vogan [71] in the Lie-
theoretic setting. First we will briefly recall contragredient duals. Let 𝑐 ∈ p be any
parameter for the rational Cherednik algebra attached to (𝑊, h). Let 𝑐 ∈ p be the
129
parameter defined by 𝑐(𝑠) = 𝑐(𝑠−1). As explained in [31, Section 3.11], there is a
natural isomorphism
𝛾 : 𝐻𝑐(𝑊, h)𝑜𝑝𝑝 → 𝐻𝑐(𝑊, h*)
acting trivially on h and h* and sending 𝑤 ↦→ 𝑤−1 for all 𝑤 ∈ 𝑊 . For any 𝑀 ∈
𝒪𝑐(𝑊, h), the restricted dual 𝑀 † :=⨁
𝑧∈C𝑀*𝑧 is naturally a 𝐻𝑐(𝑊, h)𝑜𝑝𝑝-module; by
transfer of structure along 𝛾, we regard 𝑀 † as a 𝐻𝑐(𝑊, h*)-module. We have:
Proposition 4.2.2.1. ([31, Proposition 3.32]) The assignment 𝑀 ↦→𝑀 † determines
a C-linear equivalence of categories † : 𝒪𝑐(𝑊, h) → 𝒪𝑐(𝑊, h*)𝑜𝑝𝑝.
Given a C-algebra 𝐴, let 𝐴 denote the C-algebra that is equal to 𝐴 as a ring and
as a ring is equal to the complex conjugate vector space of 𝐴. In other words, the
identity map Id : 𝐴 → 𝐴 is an isomorphism of rings and satisfies 𝑧Id(𝑎) = Id(𝑧𝑎).
Clearly 𝐴 is a C-algebra, with unit 𝜂 satisfying 𝜂(𝑧) = 𝜂(𝑧) for all 𝑧 ∈ C, where
𝜂 : C → 𝐴 is the unit map for the C-algebra 𝐴. Similarly, for any 𝐴-module 𝑀 the
complex conjugate vector space 𝑀 is naturally an 𝐴-module, and this clearly defines
an conjugate-linear equivalence of categories 𝐴-mod → 𝐴-mod.
The complex conjugate of a rational Cherednik algebra is again a rational Chered-
nik algebra. More precisely:
Lemma 4.2.2.2. Fix a nondegenerate 𝑊 -invariant Hermitian form (·, ·)h on h, and
let 𝑇 : h → h* be the conjugate-linear isomorphism introduced in Section 4.2.1. Then
the mappings 𝑦 ↦→ 𝑇𝑦 for 𝑦 ∈ h, 𝑥 ↦→ 𝑇−1𝑥 for 𝑥 ∈ h*, and 𝑤 ↦→ 𝑤 for 𝑤 ∈ 𝑊 extend
uniquely to an isomorphism of C-algebras
𝜔 : 𝐻𝑐(𝑊, h*) → 𝐻𝑐†(𝑊, h).
Proof. Regarded as a map h → h*, 𝑇 is an isomorphism of complex representations
of 𝑊 , and similarly for 𝑇−1 : h* → h. It follows that the assignments in the lemma
extend uniquely to a C-linear isomorphism C𝑊n𝑇 (h*⊕h) → C𝑊 n 𝑇 (h⊕ h*) which
determines a map C𝑊 n𝑇 (h*⊕ h) → 𝐻𝑐†(𝑊, h). As h is sent to h* and h* to h under
this map, the commutators [𝑥, 𝑥′] and [𝑦, 𝑦′] for 𝑥, 𝑥′ ∈ h* and 𝑦, 𝑦′ ∈ h are sent to 0
130
by this map. Let ⟨·, ·⟩ denote the natural pairing of h with h*. By definition of 𝑇 , we
have, for any 𝑦 ∈ h and 𝑥 ∈ h*,
⟨𝑇𝑦, 𝑇−1𝑥⟩ = (𝑇−1𝑥, 𝑦)h = (𝑦, 𝑇−1𝑥)h = ⟨𝑥, 𝑦⟩.
In particular, under the map C𝑊 n 𝑇 (h* ⊕ h) → 𝐻𝑐†(𝑊, h), for any 𝑥 ∈ h* and 𝑦 ∈ h
the image of the element
[𝑥, 𝑦] − ⟨𝑥, 𝑦⟩ +∑𝑠∈𝑆
𝑐𝑠⟨𝑥, 𝛼∨𝑠 ⟩⟨𝑦, 𝛼𝑠⟩𝑠
in 𝐻𝑐†(𝑊, h) is
[𝑇−1𝑥, 𝑇𝑦] − ⟨𝑥, 𝑦⟩ +∑𝑠∈𝑆
𝑐†𝑠⟨𝑥, 𝛼∨𝑠 ⟩⟨𝑦, 𝛼𝑠⟩𝑠
= [𝑇−1𝑥, 𝑇𝑦] − ⟨𝑇−1𝑥, 𝑇𝑦⟩ +∑𝑠∈𝑆
𝑐†𝑠⟨𝑇−1𝑥, 𝑇𝛼∨𝑠 ⟩⟨𝑇𝑦, 𝑇−1𝛼𝑠⟩𝑠.
As 𝑇𝛼∨𝑠 ∈ h* and 𝑇𝛼𝑠 ∈ h are eigenvectors for 𝑠 with nontrivial eigenvalues and
⟨𝑇𝛼∨𝑠 , 𝑇
−1𝛼𝑠⟩ = ⟨𝛼𝑠, 𝛼∨𝑠 ⟩ = 2 = 2,
the expression above is 0 in 𝐻𝑐†(𝑊, h). It follows that there is an induced map of
C-algebras 𝐻𝑐(𝑊, h*) → 𝐻𝑐†(𝑊, h), and clearly this map is an isomorphism.
Let 𝜎 : 𝐻𝑐(𝑊, h)𝑜𝑝𝑝 → 𝐻𝑐†(𝑊, h) be the isomorphism of C-algebras obtained by
compositing 𝛾 and 𝜔. Its action on generators is given by 𝜎(𝑥) = 𝑇−1𝑥 for 𝑥 ∈ h*,
𝜎(𝑦) = 𝑇𝑦 for 𝑦 ∈ h, and 𝜎(𝑤) = 𝑤−1 for 𝑤 ∈ 𝑊 . As 𝑇 depends on the choice of the
form (·, ·)h, so does 𝜎.
For any 𝑀 ∈ 𝒪𝑐(𝑊, h), the Hermitian dual 𝑀ℎ := 𝑀 † is naturally an 𝐻𝑐(𝑊, h)𝑜𝑝𝑝-
module, and by transfer of structure along 𝜎 we regard 𝑀ℎ as a 𝐻𝑐†(𝑊, h)-module.
Clearly 𝑀ℎ ∈ 𝒪𝑐†(𝑊, h), so we have:
Lemma 4.2.2.3. The assignment 𝑀 ↦→𝑀ℎ defines a conjugate-linear equivalence of
categoriesℎ : 𝒪𝑐(𝑊, h) → 𝒪𝑐†(𝑊, h)𝑜𝑝𝑝.
131
Definition 4.2.2.4. Given modules 𝑀 ∈ 𝒪𝑐(𝑊, h) and 𝑁 ∈ 𝒪𝑐†(𝑊, h), a sesquilinear
pairing
𝛽 : 𝑀 ×𝑁 → C
will be called contravariant if
𝛽(ℎ𝑚, 𝑛) = 𝛽(𝑚,𝜎(ℎ)𝑛) for all ℎ ∈ 𝐻𝑐(𝑊, h),𝑚 ∈𝑀,𝑛 ∈ 𝑁.
As 𝜎(𝑤) = 𝑤−1 for all 𝑤 ∈ 𝑊 and as 𝜎 sends the grading element of 𝐻𝑐(𝑊, h)
to the grading element of 𝐻𝑐†(𝑊, h), it follows that any contravariant pairing 𝛽 is
automatically graded and 𝑊 -invariant.
We will be particularly concerned with contravariant Hermitian forms on modules
𝑀 ∈ 𝒪𝑐(𝑊, h) for real parameters 𝑐 ∈ pR. We will use the following lemma later:
Lemma 4.2.2.5. Suppose the parameter 𝑐 ∈ p is real, i.e. 𝑐 = 𝑐†. Let 𝑀 ∈ 𝒪𝑐(𝑊, h)
be equipped with a nondegenerate 𝑊 -invariant contravariant Hermitian form 𝛽. Then
the assignment
𝑚 ↦→ 𝛽(·,𝑚)
defines an isomorphism 𝑀 ∼= 𝑀ℎ of 𝐻𝑐(𝑊, h)-modules.
Proof. That the map in question is a map of 𝐻𝑐(𝑊, h)-modules follows from the ob-
servation that the statement that 𝛽 is 𝑊 -invariant and contravariant means precisely
that 𝛽(ℎ𝑣, 𝑣′) = 𝛽(𝑣, 𝜎(ℎ)𝑣′) for all 𝑣, 𝑣′ ∈ 𝑀 and ℎ ∈ 𝐻𝑐(𝑊, h) = 𝐻𝑐†(𝑊, h). That
the map is an isomorphism follows from the nondegeneracy of 𝛽.
4.2.3 Characters and Signature Characters
Let us now briefly recall the definition of characters and signature characters.
Definition 4.2.3.1. Let 𝑀 ∈ 𝒪𝑐(𝑊, h). Then the character of 𝑀 [31, Section 3.9],
denoted ch(𝑀), is the formal series
ch(𝑀)(𝑤, 𝑡) :=∑𝑧∈C
𝑡𝑧Tr𝑀𝑧(𝑤), 𝑤 ∈ 𝑊.
132
In particular, taking 𝑤 = 1, one obtains the graded dimension of 𝑀 :
ch(𝑀)(1, 𝑡) =∑𝑧∈C
𝑡𝑧 dim𝑀𝑧.
Definition 4.2.3.2. When 𝑀 is a lowest weight module with lowest weight 𝜆, we
define the shifted character ch0(𝑀) by
ch0(𝑀)(𝑤, 𝑡) := 𝑡−ℎ𝑐(𝜆)ch(𝑀)(𝑤, 𝑡).
The character of a standard module ∆𝑐(𝜆) is given by [31, Proposition 3.27]
ch(∆𝑐(𝜆))(𝑤, 𝑡) =𝜒𝜆(𝑤)𝑡ℎ𝑐(𝜆)
deth*(1 − 𝑡𝑤).
Furthermore, it is a standard result (e.g. [31, Section 3.13]) that the set of classes
[∆𝑐(𝜆)] : 𝜆 ∈ Irr(𝑊 )
of the standard modules form a basis of the integral Grothendieck group𝐾0(𝒪𝑐(𝑊, h))
and that whenever
[𝑀 ] =∑
𝜆∈Irr(𝑊 )
𝑛𝜆[∆𝑐(𝜆)]
in 𝐾0(𝒪𝑐(𝑊, h)) we have
ch(𝑀) =∑
𝜆∈Irr(𝑊 )
𝑛𝜆ch(∆𝑐(𝜆)).
In particular, it follows from standard facts about Hilbert series that for any lowest
weight module 𝑀 we have ch0(𝑀)(1, 𝑡) = (1 − 𝑡)−𝑟𝑝𝑀(𝑡), where 𝑟 is the dimension
of support of 𝑀 and where 𝑝𝑀(𝑡) is a polynomial in 𝑡 with integer coefficients such
that 𝑝𝑀(𝑡)(1) = 0.
When 𝑀 ∈ 𝒪𝑐(𝑊, h) is equipped with a graded Hermitian form, we may similarly
define the signature character of the tuple (𝑀,𝛽):
133
Definition 4.2.3.3. Let 𝑀 ∈ 𝒪𝑐(𝑊, h) be equipped with a graded Hermitian form 𝛽.
Then the signature character sch(𝑀,𝛽) is the formal series
sch(𝑀,𝛽)(𝑡) :=∑𝑧∈C
𝑡𝑧sign(𝛽𝑧)
where, for each 𝑧 ∈ C, sign(𝛽𝑧) denotes the signature of the restriction 𝛽𝑧 of the form
𝛽 to the weight space 𝑀𝑧.
When 𝑀 ∈ 𝒪𝑐(𝑊, h) is a lowest weight module and the parameter 𝑐 is real, we
will often write sch(𝑀) rather than sch(𝑀,𝛽), where it is implicit that the Hermitian
form 𝛽 is 𝑊 -invariant, contravariant, and positive definite in the lowest weight space.
In this setting, we also define the shifted signature character:
Definition 4.2.3.4. When 𝑀 ∈ 𝒪𝑐(𝑊, h) is a lowest weight module with lowest
weight 𝜆, we define the shifted signature character sch0(𝑀) by
sch0(𝑀)(𝑡) := 𝑡−ℎ𝑐(𝜆)sch(𝑀)(𝑡).
4.2.4 Rationality of Signature Characters
The following rationality result for the shifted signature characters sch0(𝐿𝑐(𝜆)) gen-
eralizes to arbitrary finite complex reflection groups 𝑊 a corresponding result for
signature characters in type 𝐴 due to Venkateswaran [70, Corollary 1.3]:
Proposition 4.2.4.1. For any irreducible complex representation 𝜆 ∈ Irr(𝑊 ) and
real parameter 𝑐 ∈ pR, the shifted signature character sch0(𝐿𝑐(𝜆)) is of the form
sch0(𝐿𝑐(𝜆))(𝑡) = (1 − 𝑡)−𝑟𝑝𝐿𝑐(𝜆)(𝑡)
for some polynomial 𝑝𝐿𝑐(𝜆)(𝑡) with integer coefficients, where 𝑟 = dim Supp(𝐿𝑐(𝜆)).
The proof of Proposition 4.2.4.1 is an inductive argument relying on the following
lemmas:
134
Lemma 4.2.4.2. Let 𝜆 ∈ Irr(𝑊 ), and let 𝑐0, 𝑐1 ∈ pR be real parameters so that the
Jantzen filtration of ∆𝑐0(𝜆) is defined. Let 𝑐(𝑠) = 𝑐0 + 𝑠𝑐1. For sufficiently small
𝑠 > 0, we have
sch0(∆𝑐(𝑠)(𝜆))(𝑡) = sch0(∆𝑐(−𝑠)(𝜆))(𝑡) + 2𝑡−ℎ𝑐0 (𝜆)∑𝑘 odd
sch(∆𝑐0(𝜆)(𝑘), 𝛽(𝑘)𝑐0,𝜆
)(𝑡)
and
sch0(𝐿𝑐0(𝜆))(𝑡) = sch0(∆𝑐(𝑠)(𝜆))(𝑡) − 𝑡−ℎ𝑐0 (𝜆)∑𝑘≥1
sch(∆𝑐0(𝜆)(𝑘), 𝛽(𝑘)𝑐0,𝜆
)(𝑡).
Proof. This is an immediate consequence of Proposition 4.2.1.2 and the definition of
signature characters.
The following lemma (and its proof) is a reformulation for rational Cherednik
algebras of Vogan’s [71, Lemma 3.9].
Lemma 4.2.4.3. Suppose 𝑐 = 𝑐† and 𝑀 ∈ 𝒪𝑐(𝑊, h) admits a 𝑊 -invariant con-
travariant nondegenerate Hermitian form 𝛽. Suppose
[𝑀 ] =𝑛∑𝑖=1
[𝐿𝑖]
in 𝐾0(𝒪𝑐(𝑊, h)) for some simple modules 𝐿1, ..., 𝐿𝑛. Then there are 𝜖1, ..., 𝜖𝑛 ∈ ±1
such that
sch(𝑀,𝛽) =𝑛∑𝑖=1
𝜖𝑖sch(𝐿𝑖).
Proof. The proof is by induction on the length of 𝑀 . The case 𝑀 = 0 (or 𝑀 = 1) is
trivial, so suppose 𝑀 > 0 and let 𝐿 ⊂𝑀 be a simple submodule. If the restriction of
𝛽 to 𝐿 is nondegenerate, then 𝑀 = 𝐿⊕ 𝐿⊥ and 𝐿⊥ is an 𝐻𝑐(𝑊, h)-submodule of 𝑀
on which 𝛽 is nondegenerate. As all 𝑊 -invariant contravariant Hermitian forms on 𝐿
are proportional, we have sch(𝐿, 𝛽|𝐿) = ±sch(𝐿), and the claim follows by induction.
If the restriction of 𝛽 to 𝐿 is degenerate, it is 0. The inclusion 𝐿 ⊂ 𝑀 gives rise
to a surjection 𝑀ℎ → 𝐿ℎ. By Lemma 4.2.2.5, we have an isomorphism 𝑀 → 𝑀ℎ,
135
𝑚 ↦→ 𝛽(·,𝑚). The resulting composition 𝜙 : 𝑀 → 𝐿ℎ satisfies
𝜙(𝑚)(𝑙) = 𝛽(𝑙,𝑚)
for all 𝑙 ∈ 𝐿 and 𝑚 ∈ 𝑀 . As 𝛽|𝐿 = 0, we have 𝐿⊥ = ker𝜙 ⊃ 𝐿, and in particular
the form 𝛽 descends to a nondegenerate 𝑊 -invariant contravariant Hermitian form
on the subquotient 𝑁 := ker𝜙/𝐿. Now we have that 𝐿 ∼= 𝐿ℎ so
𝑀 = 𝑁 + 𝐿+ 𝐿ℎ = 𝑁 + 2𝐿
in 𝐾0(𝒪𝑐(𝑊, h)). Furthermore it follows from [71, Sublemma 3.18] that sch(𝑀) =
sch(𝑁) and hence sch(𝑀) = sch(𝑁) + sch(𝐿) − sch(𝐿), and the claim follows by
induction.
Proof of Proposition 4.2.4.1. Let 𝑐 ∈ pR. It follows from [31, Proposition 3.35] that
there are only finitely many 𝑠 ∈ [0, 2], say 𝑠1, ..., 𝑠𝑁 for some 𝑁 ≥ 0, such that
𝒪𝑠𝑐(𝑊, h) is not semisimple. Furthermore, at 𝑠 = 0 we have both that 𝒪𝑠𝑐(𝑊, h) is
semisimple and that every simple module 𝐿0(𝜆) = ∆0(𝜆) is unitary. In particular,
for all 𝜆 we have sch0(𝐿0(𝜆)) = (1 − 𝑡)−𝑙 dim𝜆 where 𝑙 = dim h, so the proposition
holds for 𝑐 = 0. Furthermore, note that the signature character sch0(∆𝑐(𝑠)(𝜆)) does
not depend on 𝑠 for those 𝑠 in a fixed interval (𝑠𝑖, 𝑠𝑖+1). So, by induction, we may
assume that the proposition holds for all 𝑠𝑐 with 𝑠 ∈ [0, 1) and we need then only
prove that it holds for all 𝑠𝑐 with 𝑠 ∈ [1, 1 + 𝛿) for some 𝛿 > 0.
For any 𝜆 ∈ Irr(𝑊 ) such that ∆𝑐0(𝜆) is simple, i.e. 𝐿𝑐0(𝜆) = ∆𝑐0(𝜆), the signa-
ture character sch0(∆(1+𝑠)𝑐0(𝜆)) does not depend on 𝑠 for |𝑠| sufficiently small. In
particular, the proposition holds for those 𝜆 ∈ Irr(𝑊 ) minimal in any highest weight
ordering ≤𝑐0 for 𝒪𝑐0(𝑊, h). By induction, we may then assume that the proposition
holds for those lowest weights 𝜇 ∈ Irr(𝑊 ) strictly lower than 𝜆 with respect to ≤𝑐0 .
Taking 𝑐1 := 𝑐0 in Lemma 4.2.4.2, it then follows from Lemmas 4.2.4.2 and 4.2.4.3
that sch0(𝐿𝑐0(𝜆)) and all sch0(∆(1+𝑠)𝑐0(𝜆)) for sufficiently small 𝑠 > 0 are of the form
(1−𝑡)−𝑙𝑝(𝑡) for some polynomial 𝑝(𝑡) with integer coefficients. As the absolute value of
136
the coefficients of the series sch0(𝐿𝑐0(𝜆)) is bounded by the coefficients of ch0(𝐿𝑐0(𝜆)),
it follows that sch0(𝐿𝑐0(𝜆)) is of the form (1 − 𝑡)−𝑟𝑝(𝑡) for some polynomial 𝑝 with
integer coefficients, as needed.
4.2.5 Asymptotic Signatures
In this section we will introduce the asymptotic signature 𝑎𝑐,𝜆 of the irreducible rep-
resentation 𝐿𝑐(𝜆), generalizing to arbitrary finite complex reflection groups the cor-
responding notion in type 𝐴 studied by Venkateswaran [70], and prove that it is a
rational number in the interval [−1, 1]. Later, in the case that Supp𝐿𝑐(𝜆) = h, we
will see in Theorem 4.4.0.1 that, in the Coxeter case, 𝑎𝑐,𝜆 is in fact a normalization of
the signature of an invariant Hermitian form on 𝐾𝑍(𝐿𝑐(𝜆)), generalizing to arbitrary
finite Coxeter groups a corresponding result in type 𝐴 due to Venkateswaran [70,
Theorem 1.4].
Lemma 4.2.5.1. Let 𝑔(𝑡) =∑∞
𝑛=0 𝑔𝑛𝑡𝑛 be a rational function regular for |𝑡| < 1 and
with a pole of maximal order at 𝑡 = 1. Let 𝑝(𝑡) and 𝑞(𝑡) be polynomials with 𝑞(1) = 0,
and let 𝑠(𝑡) = 𝑝(𝑡)𝑔(𝑡) =∑∞
𝑛=0 𝑠𝑛𝑡𝑛 and 𝑑(𝑡) = 𝑞(𝑡)𝑔(𝑡) =
∑∞𝑛=0 𝑑𝑛𝑡
𝑛. Then the limits
lim𝑡→1−
𝑠(𝑡)
𝑑(𝑡), lim
𝑁→∞
∑𝑛≤𝑁 𝑠𝑛∑𝑛≤𝑁 𝑑𝑛
both exist and equal 𝑝(1)/𝑞(1).
Proof. That the first limit exists and equals 𝑝(1)/𝑞(1) is clear. Let 𝑟 be the order of
the pole of 𝑔 at 𝑡 = 1. We have
∞∑𝑁=0
(∑𝑛≤𝑁
𝑠𝑛
)𝑡𝑁 = 𝑝(𝑡)𝑔(𝑡)/(1 − 𝑡) = (𝑝(𝑡) − 𝑝(1))𝑔(𝑡)/(1 − 𝑡) + 𝑝(1)𝑔(𝑡)/(1 − 𝑡).
As 𝑔(𝑡)/(1 − 𝑡) has a pole of order 𝑟 + 1 at 𝑡 = 1, greater than the order of any
other pole of 𝑔(𝑡)/(1 − 𝑡) or of any pole of (𝑝(𝑡) − 𝑝(1))𝑔(𝑡)/(1 − 𝑡), it follows from a
consideration of partial fractions that lim𝑁→∞(∑
𝑛≤𝑁 𝑠𝑛)/𝑁 𝑟 = 𝑝(1)𝐶, where 𝐶 = 0
is a nonzero constant depending only on the rational function 𝑔(𝑡). Similarly, we have
137
lim𝑁→∞(∑
𝑛≤𝑁 𝑑𝑛)/𝑁 𝑟 = 𝑞(1)𝐶, and the claim follows.
We will use Lemma 4.2.5.1 in the special cases that 𝑔(𝑡) = (1− 𝑡)−𝑙 when working
signature characters and 𝑔(𝑡) =∏𝑙
𝑖=1(1 − 𝑡𝑑𝑖)−1, where 𝑑1, ..., 𝑑𝑙 are the fundamental
degrees of the reflection group𝑊 , when working with the isotypic signature characters
introduced in Section 4.2.6.
Lemma 4.2.5.2. Let 𝑐 ∈ pR be a real parameter and let 𝜆 ∈ Irr(𝑊 ), so that the
irreducible lowest weight module 𝐿𝑐(𝜆) admits a 𝑊 -invariant contravariant nonde-
generate Hermitian form 𝛽𝑐,𝜆, normalized to be positive definite on 𝜆. For any 𝑛 ≥ 0,
let 𝛽≤𝑛𝑐,𝜆 denote the restriction of 𝛽 to the space 𝐿𝑐(𝜆)≤𝑛 :=
⨁𝑘≤𝑛 𝐿𝑐(𝜆)[𝑘]. Then the
limits
lim𝑛→∞
sign(𝛽≤𝑛𝑐,𝜆 )
dim𝐿𝑐(𝜆)≤𝑛, lim
𝑡→1−
sch0(𝐿𝑐(𝜆))(𝑡)
ch0(𝐿𝑐(𝜆))(1, 𝑡)
exist and are equal to the same rational number 𝑎𝑐,𝜆 ∈ [−1, 1]. If
𝑟 := dim Supp(𝐿𝑐(𝜆)) > 0,
then the limit
lim𝑛→∞
sign(𝛽𝑐,𝜆[𝑛])
dim𝐿𝑐(𝜆)[𝑛]
also exists and equals 𝑎𝑐,𝜆.
Proof. In the case 𝑟 = 0 the claim is clear, so we may assume 𝑟 > 0. Using Propo-
sition 4.2.4.1, write ch0(𝐿𝑐(𝜆))(1, 𝑡) =∑∞
𝑛=0 𝑑𝑛𝑡𝑛 = (1 − 𝑡)−𝑟𝑞(𝑡) and sch0(𝐿𝑐(𝜆)) =∑∞
𝑛=0 𝑠𝑛𝑡𝑛 = (1 − 𝑡)−𝑟𝑝(𝑡), where 𝑝(𝑡) and 𝑞(𝑡) are polynomials with integer coeffi-
cients satisfying 𝑞(1) = 0, where we use the series expansion about 0 for (1 − 𝑡)−𝑟.
In particular, Lemma 4.2.5.1 applies, and the first two limits in the lemma statement
exist and equal 𝑝(1)/𝑞(1) ∈ Q. As we clearly have |sign(𝛽≤𝑛𝑐,𝜆 )| ≤ dim𝐿𝑐(𝜆)≤𝑛, we also
have 𝑝(1)/𝑞(1) ∈ [−1, 1], giving the first claim. The final claim follows similarly, not-
ing that for 𝑛 >> 0 we have that sign(𝛽𝑐,𝜆[𝑛]) and dim𝐿𝑐(𝜆)[𝑛] are polynomials in 𝑛
of degree at most 𝑟 and with coefficients of 𝑛𝑟 given by 𝑞(1)/(𝑟−1)! and 𝑝(1)/(𝑟−1)!,
respectively.
138
Definition 4.2.5.3. For 𝑐 ∈ pR, the rational number 𝑎𝑐,𝜆 ∈ [−1, 1] appearing in
Lemma 4.2.5.2 is the asymptotic signature of 𝐿𝑐(𝜆). If 𝑎𝑐,𝜆 = ±1, we say 𝐿𝑐(𝜆) is
quasi-unitary.
4.2.6 Isotypic Signature Characters
As contravariant forms are not only graded but also 𝑊 -invariant, it is natural to
consider the isotypic signature characters recording the signatures of contravariant
forms both within graded components and also within 𝑊 -isotypic components:
Definition 4.2.6.1. Let 𝑐 ∈ pR be a real parameter, let 𝛽 be a contravariant Her-
mitian form on the module 𝑀 ∈ 𝒪𝑐(𝑊, h), and for all 𝑧 ∈ C and 𝜋 ∈ Irr(𝑊 ) let
𝑀𝜋𝑧 ⊂ 𝑀 denote the 𝜋-isotypic component of the homogeneous degree 𝑧 part of 𝑀
and let 𝛽𝜋𝑧 = 𝛽|𝑀𝜋𝑧. The 𝜋-isotypic signature character sch𝜋(𝑀,𝛽) of 𝑀 with respect
to 𝛽 is the formal series
sch𝜋(𝑀,𝛽)(𝑡) :=∑𝑧∈C
sign(𝛽𝜋𝑧 )𝑡𝑧.
Clearly, we have sch(𝑀,𝛽) =∑
𝜋∈Irr(𝑊 ) sch𝜋(𝑀,𝛽). When 𝑀 is lowest weight, we
define sch𝜋0 (𝑀)(𝑡) ∈ Z[[𝑡]] completely analogously to sch0(𝑀). Similarly, let ch𝜋(𝑀)
denote the graded dimension of 𝑀𝜋, and when 𝑀 is lowest weight with lowest weight
𝜆 let ch𝜋0 (𝑀)(𝑡) = 𝑡−ℎ𝑐(𝜆)ch𝜋(𝑀)(𝑡) ∈ Z≥0[[𝑡]].
The identification of any standard module ∆𝑐(𝜆) with C[h] ⊗ 𝜆 used in Section
4.2.1 respects the 𝑊 -action. In particular, in the setting considered in that section,
each isotypic subspace ∆𝑐(𝜆)𝜋 is equipped with a Jantzen filtration ∆𝑐(𝜆)𝜋,≥𝑘𝑘≥0,
and we have ∆𝑐(𝜆)≥𝑘 =⨁
𝜋∈Irr(𝑊 ) ∆𝑐(𝜆)𝜋,≥𝑘. Furthermore, the wall-crossing formulas
in Lemma 4.2.4.2 and the decomposition in Lemma 4.2.4.3 of an arbitrary signature
character sch(𝑀,𝛽) in terms of signature characters sch(𝐿𝑖) of irreducible represen-
tations, and their proofs, have direct analogues for the isotypic signatures characters
sch𝜋.
Let 𝑙 = dim h, let 𝑑𝑖𝑙𝑖=1 denote the fundamental degrees of 𝑊 , and let C[h]𝑐𝑜𝑊 :=
C[h]/C[h](C[h]𝑊 )+ be the coinvariant algebra. Recall that C[h]𝑐𝑜𝑊 is graded and is
139
isomorphic to C𝑊 as a C𝑊 -module and that the graded dimension of C[h]𝑊 is given
by∏𝑙
𝑖=1(1−𝑡𝑑𝑖)−1. For any 𝜆, 𝜋 ∈ Irr(𝑊 ), let 𝜃𝜋𝜆 ∈ Z≥0[𝑡] denote the graded dimension
of the 𝜋-isotypic subspace of 𝜆⊗ C[h]𝑐𝑜𝑊 .
Lemma 4.2.6.2. For any 𝜆, 𝜋 ∈ Irr(𝑊 ), we have:
(1) 𝜃𝜋𝜆(1) = (dim𝜆)(dim𝜋)2
(2) ch𝜋0 (∆0(𝜆)) = sch𝜋0 (∆0(𝜆)) = 𝜃𝜋𝜆(𝑡)/∏𝑙
𝑖=1(1 − 𝑡𝑑𝑖).
(3)
lim𝑛→∞
dim ∆0(𝜆)𝜋,≤𝑛
dim ∆0(𝜆)≤𝑛= lim
𝑡→1−
ch𝜋0 (∆0(𝜆))(𝑡)
ch0(∆0(𝜆))(1, 𝑡)=
(dim𝜋)2
|𝑊 |.
Proof. 𝜃𝜋𝜆(1) is the dimension of the 𝜋-isotypic subspace of 𝜆⊗C[h]𝑐𝑜𝑊 ; as C[h]𝑐𝑜𝑊 ∼=
C𝑊 as a C𝑊 -module and as 𝜆⊗ C𝑊 ∼= C𝑊⊕ dim𝜆, the first statement follows. The
second statement follows from the fact that C[h] ∼= C[h]𝑐𝑜𝑊 ⊗ C[h]𝑊 as a graded
C𝑊 -module. The third statement follows from the first two statements and from
Lemma 4.2.5.1, using the facts that ch0(∆0(𝜆))(1, 𝑡) = (dim𝜆)𝑃𝑊 (𝑡)/∏𝑙
𝑖=1(1 − 𝑡𝑑𝑖),
where 𝑃𝑊 (𝑡) is the Poincaré polynomial of 𝑊 , and 𝑃𝑊 (1) = |𝑊 |.
Note that Lemma 4.2.4.2 holds for the isotypic signature characters sch𝜋 as well,
by the same proof as for the usual signature characters sch. As the signs 𝜖𝑖 appearing
in Lemma 4.2.4.3 have no dependence on the irreducible representation 𝜋, it follows
from Lemma 4.2.6.2 and the proof of Proposition 4.2.4.1 that we have:
Lemma 4.2.6.3. For any 𝑐 ∈ pR, there exists a collection of polynomials 𝑛𝜆,𝜇𝑐 (𝑡) ∈
Z[𝑡] : 𝜆, 𝜇 ∈ Irr(𝑊 ) such that for every 𝜆, 𝜋 ∈ Irr(𝑊 ) we have
sch𝜋0 (𝐿𝑐(𝜆))(𝑡) =∑
𝜇∈Irr(𝑊 )
𝑛𝜆,𝜇𝑐 (𝑡)sch𝜋0 (∆0(𝜇))(𝑡)
=𝑙∏
𝑖=1
(1 − 𝑡𝑑𝑖)−1∑
𝜇∈Irr(𝑊 )
𝑛𝜆,𝜇𝑐 (𝑡)𝜃𝜋𝜇(𝑡).
In particular, sch𝜋0 (𝐿𝑐(𝜆))(𝑡) is a rational function of 𝑡.
Corollary 4.2.6.4. In the setting of Lemma 4.2.6.3, the rational function
sch𝜋0 (𝐿𝑐(𝜆))(𝑡)
140
has a pole of order at most 𝑟 := dim Supp(𝐿𝑐(𝜆)) at 𝑡 = 1.
Proof. By Lemma 4.2.6.3, we see that sch𝜋0 (𝐿𝑐(𝜆))(𝑡) is absolutely convergent for
|𝑡| < 1, and by a comparison of coefficients we see that the rational function
sch𝜋0 (𝐿𝑐(𝜆))(𝑡)/ch0(𝐿𝑐(𝜆))(1, 𝑡)
takes values in [−1, 1] on the interval [0, 1). As ch0(𝐿𝑐(𝜆))(1, 𝑡) has a pole of order 𝑟
at 𝑡 = 1, the claim follows.
The following proposition allows the asymptotic signature 𝑎𝑐,𝜆 of 𝐿𝑐(𝜆) to be
computed in any isotypic component when 𝐿𝑐(𝜆) has full support. Considering the
case when 𝐿𝑐(𝜆) is finite-dimensional, note that this need not be true when 𝐿𝑐(𝜆) has
proper support.
Proposition 4.2.6.5. Let 𝑐 ∈ pR, 𝜆 ∈ Irr(𝑊 ), and suppose 𝐿𝑐(𝜆) has full support.
Then for all 𝜋 ∈ Irr(𝑊 ) the limit
lim𝑛→∞
sign(𝛽𝜋,≤𝑛𝑐,𝜆 )
dim𝐿𝑐(𝜆)𝜋,≤𝑛
exists and equals 𝑎𝑐,𝜆. In particular, this limit is independent of 𝜋.
Proof. As we have
𝑎𝑐,𝜆 = lim𝑛→∞
sign(𝛽≤𝑛𝑐,𝜆 )
dim𝐿𝑐(𝜆)≤𝑛= lim
𝑛→∞
∑𝜋∈Irr(𝑊 ) sign(𝛽𝜋,≤𝑛𝑐,𝜆 )∑
𝜋∈Irr(𝑊 ) dim𝐿𝑐(𝜆)𝜋,≤𝑛
it suffices to show that the limit in the proposition statement exists and is inde-
pendent of 𝜋. As 𝐿𝑐(𝜆) has full support, considering the decomposition of [𝐿𝑐(𝜆)]
in 𝐾0(𝒪𝑐(𝑊, h)) in terms of the classes of standard modules [∆𝑐(𝜇)] for various
𝜇 ∈ Irr(𝑊 ), we have
lim𝑛→∞
dim𝐿𝑐(𝜆)𝜋,≤𝑛
dim𝐿𝑐(𝜆)≤𝑛= lim
𝑡→1−
ch𝜋0 (𝐿𝑐(𝜆))(𝑡)
ch0(𝐿𝑐(𝜆))(1, 𝑡)=
(dim𝜋)2
|𝑊 |(4.2.1)
141
where the first equality follows from Lemma 4.2.5.1 and the second equality follows
from Lemma 4.2.6.2(3) and the fact that 𝐿𝑐(𝜆) has full support in h. By equation
(4.2.1) and Lemmas 4.2.5.1 and 4.2.6.2 we have
lim𝑛→∞
sign(𝛽𝜋,≤𝑛𝑐,𝜆 )
dim𝐿𝑐(𝜆)𝜋,≤𝑛
=|𝑊 |
(dim𝜋)2lim𝑛→∞
sign(𝛽𝜋,≤𝑛𝑐,𝜆 )
dim𝐿𝑐(𝜆)≤𝑛
=|𝑊 |
(dim𝜋)2lim𝑡→1−
sch𝜋0 (𝐿𝑐(𝜆))(𝑡)
ch0(𝐿𝑐(𝜆))(1, 𝑡)
=|𝑊 |
(dim𝜋)2lim𝑡→1−
∑𝜇∈Irr(𝑊 ) 𝑛
𝜆,𝜇𝑐 (𝑡)𝜃𝜋𝜇(𝑡)/
∏𝑙𝑖=1(1 − 𝑡𝑑𝑖)
ch0(𝐿𝑐(𝜆))(1, 𝑡)
= |𝑊 | lim𝑡→1−
∑𝜇∈Irr(𝑊 ) 𝑛
𝜆,𝜇𝑐 (1) dim𝜇
ch0(𝐿𝑐(𝜆))(1, 𝑡)∏𝑙
𝑖=1(1 − 𝑡𝑑𝑖),
which is visibly independent of 𝜋, as needed. Note that this final limit exists because
𝐿𝑐(𝜆) has full support in h, so ch0(𝐿𝑐(𝜆))(1, 𝑡) as a pole of order 𝑙 at 𝑡 = 1.
4.3 The Dunkl Weight Function
4.3.1 The Contravariant Pairing and Contravariant Form
As in the previous section, let 𝑊 be a finite complex reflection group with reflection
representation h. Fix a positive-definite 𝑊 -invariant inner product on hR, and let
(·, ·)h be its unique extension to a 𝑊 -invariant positive-definite Hermitian form on h.
Let 𝑇 : h → h*, 𝑇 (𝑦)(𝑥) = (𝑥, 𝑦)h, be the conjugate-linear 𝑊 -invariant isomorphism
introduced in Section 4.2.1, giving rise for any parameter 𝑐 ∈ p to the isomorphism 𝜎 :
𝐻𝑐(𝑊, h)𝑜𝑝𝑝 → 𝐻𝑐†(𝑊, h) of C-algebras and conjugate-linear equivalence of categories
ℎ : 𝒪𝑐(𝑊, h) → 𝒪𝑐†(𝑊, h)𝑜𝑝𝑝
introduced in Section 4.2.2.
142
Let 𝜆 ∈ Irr(𝑊 ) be an irreducible representation of 𝑊 , and fix a positive-definite
𝑊 -invariant Hermitian form (·, ·)𝜆 on 𝜆. For any 𝑐 ∈ p, the 𝑊 -invariant isomorphism
𝜙 : 𝜆→ 𝜆* given by 𝜙(𝑣)(𝑣′) = (𝑣′, 𝑣) between the lowest weight spaces of ∆𝑐†(𝜆) and
∆𝑐(𝜆)ℎ extends uniquely to an𝐻𝑐†(𝑊, h)-homomorphism 𝜙 : ∆𝑐†(𝜆) → ∆𝑐(𝜆)ℎ. When
it is convenient, we will use the notation 𝑣†2𝑣1 to denote the inner product (𝑣1, 𝑣2)𝜆
for 𝑣1, 𝑣2 ∈ 𝜆, borrowing notation from the standard inner product on C𝑁 . Similarly,
we will use the notation 𝐴† to denote the adjoint of an operator 𝐴 ∈ EndC(𝜆) with
respect to the form (·, ·)𝜆.
Definition 4.3.1.1. For 𝑐 ∈ p and 𝜆 ∈ Irr(𝑊 ), let 𝛽𝑐,𝜆 be the sesquilinear pairing
𝛽𝑐,𝜆 : ∆𝑐(𝜆) × ∆𝑐†(𝜆) → C
given by
𝛽𝑐,𝜆(𝑣, 𝑣′) = 𝜙(𝑣′)(𝑣).
Remark 4.3.1.2. 𝛽𝑐,𝜆 depends on the choice of forms (·, ·)h and (·, ·)𝜆 and is therefore
determined by 𝑐 ∈ p and 𝜆 ∈ Irr(𝑊 ) only up to a positive real multiple.
Remark 4.3.1.3. When the parameter 𝑐 ∈ p is real, i.e. 𝑐 ∈ pR and 𝑐 = 𝑐†, the
pairing 𝛽𝑐,𝜆 is precisely the contravariant Hermitian form on ∆𝑐(𝜆) considered in
[10] and [33, Definition 4.5]. While this case is the primary case of interest, we
will consider the more general setting of complex 𝑐 to show that the associated Dunkl
weight function, introduced later, is defined and holomorphic for all 𝑐 ∈ p.
The following proposition is standard (for 𝑐 ∈ pR it is a reformulation of [33,
Proposition 2.2]):
Proposition 4.3.1.4.
(i) 𝛽𝑐,𝜆 is the unique sesquilinear form 𝛽𝑐,𝜆 : ∆𝑐(𝜆) × ∆𝑐†(𝜆) → C that is:
(a) normalized: 𝛽𝑐,𝜆|𝜆 = (·, ·)𝜆(b) contravariant: 𝛽𝑐,𝜆(ℎ𝑣, 𝑣
′) = 𝛽𝑐,𝜆(𝑣, 𝜎(ℎ)𝑣′) for all ℎ ∈ 𝐻𝑐(𝑊, h), 𝑣 ∈
∆𝑐(𝜆), and 𝑣′ ∈ ∆𝑐†(𝜆),
143
(ii) Any contravariant form 𝛽 : ∆𝑐(𝜆) × ∆𝑐†(𝜆) → C is proportional to 𝛽𝑐,𝜆.
(iii) 𝛽𝑐,𝜆 is graded, 𝑊 -invariant, and satisfies 𝛽𝑐,𝜆 = 𝛽†𝑐†,𝜆
.
(iv) 𝛽𝑐,𝜆 descends to a nondegenerate pairing 𝛽𝑐,𝜆 : 𝐿𝑐(𝜆) × 𝐿𝑐†(𝜆) → C.
4.3.2 The Gaussian Pairing and Gaussian Inner Product
For the remainder of this chapter, let 𝑊 be a finite real reflection group (equivalently,
a finite Coxeter group). Let hR denote its real reflection representation, let h = C⊗RhR
denote the complexified reflection representation, let 𝛼𝑠𝑠∈𝑆 ⊂ h*R be a system of
positive roots for 𝑊 , let 𝛼∨𝑠 𝑠∈𝑆 ⊂ hR be the associated system of positive coroots,
and fix an orthonormal basis 𝑦1, ..., 𝑦𝑙 of hR with associated dual basis 𝑥1, ..., 𝑥𝑙 ∈ h*R.
In this setting, 𝐻𝑐(𝑊, h) contains an internal 𝑊 -invariant sl2-triple e, f ,h given by
h =∑𝑖
𝑥𝑖𝑦𝑖+1
2dim h−
∑𝑠∈𝑆
𝑐𝑠𝑠 =∑𝑖
(𝑥𝑖𝑦𝑖+𝑦𝑖𝑥𝑖)/2, e := −1
2
∑𝑖
𝑥2𝑖 f :=1
2
∑𝑖
𝑦2𝑖 .
The elements e, f ,h do not depend on the choice of orthonormal basis 𝑦1, ..., 𝑦𝑙 ∈ hR.
Note that the operator f acts locally nilpotently on any lowest weight module. In
particular, following [10] and [33, Definition 4.5], we make the following definition:
Definition 4.3.2.1. The Gaussian pairing 𝛾𝑐,𝜆 is the sesquilinear pairing
𝛾𝑐,𝜆 : ∆𝑐(𝜆) × ∆𝑐†(𝜆) → C
defined by
𝛾𝑐,𝜆(𝑃,𝑄) = 𝛽𝑐,𝜆(exp(f)𝑃, exp(f)𝑄).
When 𝑐 ∈ pR, 𝛾𝑐,𝜆 is a Hermitian form on ∆𝑐(𝜆) and will be called the Gaussian inner
product.
Remark 4.3.2.2. For 𝑐 ∈ pR, the study of the signatures of the Hermitian forms 𝛽𝑐,𝜆
and 𝛾𝑐,𝜆 is equivalent; we have sign(𝛽≤𝑛𝑐,𝜆 ) = sign(𝛾≤𝑛𝑐,𝜆 ) for all 𝑛 ≥ 0, where 𝛽≤𝑛
𝑐,𝜆 and
𝛾≤𝑛𝑐,𝜆 denote the restrictions of 𝛽𝑐,𝜆 and 𝛾𝑐,𝜆, respectively, to ∆𝑐(𝜆)≤𝑛. In particular,
the asymptotic signature 𝑎𝑐,𝜆 can be equally well computed using the form 𝛾𝑐,𝜆. As
144
we will see, the advantage of the form 𝛾𝑐,𝜆 is that h*R acts on ∆𝑐(𝜆) by self-adjoint
operators, making an integral representation of this form possible and allowing for
analytic techniques to be used to study the signatures.
The following proposition generalizes [33, Proposition 4.6] to complex 𝑐 ∈ p and
provides a useful characterization of 𝛾𝑐,𝜆:
Proposition 4.3.2.3.
(i) The pairing 𝛾𝑐,𝜆 satisfies:
𝛾𝑐,𝜆(𝑥𝑃,𝑄) = 𝛾𝑐,𝜆(𝑃, 𝑥𝑄) for all 𝑥 ∈ h*R, 𝑃 ∈ ∆𝑐(𝜆), 𝑄 ∈ ∆𝑐†(𝜆).
(ii) Up to scaling, 𝛾𝑐,𝜆 is the unique 𝑊 -invariant sesquilinear pairing ∆𝑐(𝜆) ×
∆𝑐†(𝜆) → C satisfying
𝛾𝑐,𝜆((−𝑦 + 𝑇𝑦)𝑃,𝑄) = 𝛾𝑐,𝜆(𝑃, 𝑦𝑄) for all 𝑦 ∈ hR, 𝑃 ∈ ∆𝑐(𝜆), 𝑄 ∈ ∆𝑐†(𝜆)
and
𝛾𝑐,𝜆(𝑦𝑃,𝑄) = 𝛾𝑐,𝜆(𝑃, (−𝑦 + 𝑇𝑦)𝑄) for all 𝑦 ∈ hR, 𝑃 ∈ ∆𝑐(𝜆), 𝑄 ∈ ∆𝑐†(𝜆).
𝛾𝑐,𝜆 is determined among such forms by the normalization condition 𝛾𝑐,𝜆|𝜆 = (·, ·)𝜆.
(iii) Up to scaling, 𝛾𝑐,𝜆 is the unique 𝑊 -invariant sesquilinear pairing ∆𝑐(𝜆) ×
∆𝑐†(𝜆) → C satisfying
𝛾𝑐,𝜆(𝑥𝑃,𝑄) = 𝛾𝑐,𝜆(𝑃, 𝑥𝑄) for all 𝑥 ∈ h*R, 𝑃 ∈ ∆𝑐(𝜆), 𝑄 ∈ ∆𝑐†(𝜆)
and
𝛾𝑐,𝜆((−𝑦 + 𝑇𝑦)𝑃, 𝑣) = 0 for all 𝑦 ∈ hR, 𝑃 ∈ ∆𝑐(𝜆), 𝑣 ∈ 𝜆.
𝛾𝑐,𝜆 is determined among such forms by the normalization condition 𝛾𝑐,𝜆|𝜆 = (·, ·)𝜆.
(iv) 𝛾𝑐,𝜆 = 𝛾†𝑐†,𝜆
.
Proof. Parts (i) and (ii) are proved using Proposition 4.3.1.4 and exactly the same
145
calculations and arguments appearing in the proof of [33, Proposition 4.6]. To prove
(iii), it suffices to show that the properties in (iii) imply the properties in (ii). Let
𝛾 be a pairing as in (iii). As h*R acts by operators self-adjoint with respect to 𝛾, the
two equations appearing in (ii) are equivalent because 𝑇𝑦 ∈ h*R is self-adjoint, so it
suffices to prove that
𝛾((−𝑦 + 𝑇𝑦)𝑃,𝑄) = 𝛾(𝑃, 𝑦𝑄) (4.3.1)
for all 𝑦 ∈ hR, 𝑃 ∈ ∆𝑐(𝜆), and 𝑄 ∈ ∆𝑐†(𝜆). We will prove equation (4.3.1) by
induction on deg𝑄. As 𝑦𝑣 = 0 for all 𝑦 ∈ hR and 𝑣 ∈ 𝜆, equation (4.3.1) holds for all
𝑦 ∈ hR, 𝑃 ∈ ∆𝑐(𝜆), and 𝑄 ∈ 𝜆 by the second property in (iii), establishing the base
case deg𝑄 = 0. Now suppose, for some 𝑁 ≥ 0, that equation (4.3.1) holds for all 𝑦 ∈
hR, 𝑃 ∈ ∆𝑐(𝜆), and 𝑄 ∈ ∆𝑐†(𝜆) with deg𝑄 ≤ 𝑁 . For any 𝑥 ∈ h*R, 𝑦 ∈ hR, 𝑃 ∈ ∆𝑐(𝜆),
and 𝑄 ∈ ∆𝑐†(𝜆) we have
𝛾((−𝑦 + 𝑇𝑦)𝑃, 𝑥𝑄) − 𝛾(𝑃, 𝑦𝑥𝑄)
= 𝛾(𝑥(−𝑦 + 𝑇𝑦)𝑃,𝑄) − 𝛾
(𝑃,
(𝑥𝑦 + 𝑥(𝑦) −
∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)𝛼∨𝑠 (𝑥)𝑠
)𝑄
)
= 𝛾(𝑥(−𝑦 + 𝑇𝑦)𝑃,𝑄) − 𝛾
(((−𝑦 + 𝑇𝑦)𝑥+ 𝑥(𝑦) −
∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)𝛼∨𝑠 (𝑥)𝑠
)𝑃,𝑄
)= 0
where we use the commutation relation for [𝑦, 𝑥] in 𝐻𝑐†(𝑊, h) and 𝐻𝑐(𝑊, h) in the
first and last equality, respectively, and the fact that [𝑥, 𝑇𝑦] = 0, establishing the
inductive step.
Statement (iv) follows immediately from 𝛽𝑐,𝜆 = 𝛽†𝑐†,𝜆
.
4.3.3 Dunkl Weight Function: Characterization and Proper-
ties
The remainder of Section 4.3 is dedicated to proving the following main theorem:
146
Theorem 4.3.3.1. For any finite Coxeter group 𝑊 and irreducible representation
𝜆 ∈ Irr(𝑊 ), there is a unique family 𝐾𝑐,𝜆, holomorphic in 𝑐 ∈ p, of EndC(𝜆)-valued
tempered distributions on hR such that the following integral representation of the
Gaussian pairing 𝛾𝑐,𝜆 holds for all 𝑐 ∈ p:
𝛾𝑐,𝜆(𝑃,𝑄) =
∫hR
𝑄(𝑥)†𝐾𝑐,𝜆(𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥 for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆,
where we make the standard identification of ∆𝑐(𝜆) and ∆𝑐†(𝜆) with C[h] ⊗ 𝜆. Fur-
thermore, 𝐾𝑐,𝜆 satisfies the additional properties:
(i) For all 𝑀 > 0 there exists an integer 𝑁 ≥ 0, which may be taken to be 0 for
𝑀 sufficiently small, such that for 𝑐 ∈ p with |𝑐𝑠| < 𝑀 for all 𝑠 ∈ 𝑆 the distribution
𝛿𝑁𝐾𝑐,𝜆, where 𝛿 :=∏
𝑠∈𝑆 𝛼𝑠 ∈ C[h] is the discriminant element, is given by integration
against an analytic function on hR,𝑟𝑒𝑔 that is locally integrable over hR.
(ii) For any 𝑥 ∈ hR,𝑟𝑒𝑔 the operator 𝐾𝑐,𝜆(𝑥) ∈ EndC(𝜆) determines a 𝐵𝑊 -invariant
sesquilinear pairing
𝐾𝑍𝑥(∆𝑐(𝜆)) ×𝐾𝑍𝑥(∆𝑐†(𝜆)) → C
by the formula (𝑣1, 𝑣2) = 𝑣†2𝐾𝑐,𝜆(𝑥)𝑣1. When 𝑐 ∈ pR, i.e. 𝑐 = 𝑐†, this pairing is a
Hermitian form.
Remark 4.3.3.2. The uniqueness of 𝐾𝑐,𝜆 is immediate from the standard fact that
the subspace C[h]𝑒−|𝑥|2/2 is dense in the space of complex-valued Schwartz functions
S (hR) on hR.
Remark 4.3.3.3. The existence of the distribution 𝐾𝑐,𝜆 was shown in the case that
𝑊 is a dihedral group and 𝑐 is real and small (for any 𝜆) by Dunkl [19]. The case in
which 𝜆 is 1-dimensional is much simpler than the case of general 𝜆 and was used by
Etingof in [28] to study the support of the irreducible representation 𝐿𝑐(triv). Related
results in the trigonometric case in type 𝐴 appear in [22, 23, 24].
In preparation for the proof of Theorem 4.3.3.1, in the remainder of Section 4.3.3
we will now establish several necessary properties and an additional characterization
of the distribution𝐾𝑐,𝜆. The proof of Theorem 4.3.3.1 will then proceed in three steps:
147
in Section 4.3.4 we will establish the existence of 𝐾𝑐,𝜆 for small 𝑐 and show that in
this case it is given by integration against a locally 𝐿1 function; in Section 4.3.5 we
will produce an analytic continuation for all 𝑐 ∈ p of this function over hR,𝑟𝑒𝑔; and in
Section 4.3.6 we will show that these functions extend naturally to distributions on
all of hR, completing the proof of Theorem 4.3.3.1.
An element 𝑦 ∈ h ⊂ 𝐻𝑐(𝑊, h) acts in the representation ∆𝑐(𝜆) = C[h] ⊗ 𝜆 by the
operator
𝐷𝑦 = 𝜕𝑦 −∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(1 − 𝑠) ⊗ 𝑠.
This operator 𝐷𝑦 acts naturally as a continuous operator on the Schwartz space
S (hR), and we will use the notation 𝐷𝑦 for this continuous operator as well, with
the representation 𝜆 and parameter 𝑐 implicit and suppressed from the notation. We
may equivalently view any tempered distribution 𝐾 on hR with values in EndC(𝜆) as
a continuous operator
𝐾 : S (hR) ⊗ 𝜆→ 𝜆.
In particular, for any such distribution 𝐾, the composition 𝐾 ∘𝐷𝑦 is defined and is
another such distribution. With this in mind, it is now straightforward to translate
the conditions in Proposition 4.3.2.3(iii) into a convenient characterization of the
Dunkl weight function 𝐾𝑐,𝜆:
Proposition 4.3.3.4. Let 𝐾 be an EndC(𝜆)-valued tempered distribution on hR, and
let 𝛾 be the sesquilinear form on C[h] ⊗ 𝜆 defined by
𝛾(𝑃,𝑄) =
∫hR
𝑄(𝑥)†𝐾(𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥.
Then 𝛾 = 𝛾𝑐,𝜆 if and only if the following hold:
(i) normalization: ∫hR
𝐾(𝑥)𝑒−|𝑥|2/2𝑑𝑥 = Id𝜆
(ii) 𝑊 -invariance: 𝑤𝐾(𝑤−1𝑥)𝑤−1 = 𝐾(𝑥) for all 𝑤 ∈ 𝑊
(iii) annihilation by Dunkl operators: 𝐾 ∘𝐷𝑦 = 0 for all 𝑦 ∈ hR.
148
Proof. It is immediate that the normalization condition (i) above and the normal-
ization condition 𝛾|𝜆 = (·, ·)𝜆 are equivalent and that the 𝑊 -invariance condition
(ii) above is equivalent to the 𝑊 -invariance of the form 𝛾. It is also immediate
that 𝛾(𝑥𝑃,𝑄) = 𝛾(𝑃, 𝑥𝑄) for all 𝑥 ∈ hR and 𝑃,𝑄 ∈ C[h] ⊗ 𝜆, so it remains to
show that condition (iii) is equivalent to the condition 𝛾((−𝑦𝑖 + 𝑥𝑖)𝑃, 𝑣) = 0 for all
𝑃 ∈ C[h] ⊗ 𝜆, 𝑣 ∈ 𝜆, and 𝑖 = 1, ..., 𝑙 = dim h. As 𝑒−|𝑥|2/2 is 𝑊 -invariant we have
𝐷𝑦𝑖(𝑃 (𝑥)𝑒−|𝑥|2/2) = (𝐷𝑦𝑖𝑃 (𝑥) − 𝑥𝑖)𝑒−|𝑥|2/2, and a direct calculation then gives
𝛾((−𝑦𝑖 + 𝑥𝑖)𝑃, 𝑣) =
∫hR
𝑣†𝐾(𝑥)(−𝑦𝑖 + 𝑥𝑖)(𝑃 (𝑥))𝑒−|𝑥|2/2𝑑𝑥
=
∫hR
𝑣†(𝐾 ∘𝐷𝑦𝑖)(𝑃 (𝑥)𝑒−|𝑥|2/2)𝑑𝑥.
As C[h]𝑒−|𝑥|2/2 is dense in S (hR), we see that 𝛾((−𝑦𝑖+𝑥𝑖)𝑃, 𝑣) = 0 for all 𝑃 ∈ C[h]⊗𝜆
and 𝑣 ∈ 𝜆 if and only if 𝐾 ∘𝐷𝑦𝑖 = 0, as needed.
Proposition 4.3.3.5. Any family of tempered distributions 𝐾𝑐,𝜆 as appearing in The-
orem 4.3.3.1 satisfies the equation 𝐾𝑐,𝜆 = 𝐾†𝑐†,𝜆
. In particular, 𝐾𝑐,𝜆 takes values in
Hermitian forms on 𝜆 when 𝑐 ∈ pR.
Proof. Immediate from the equality 𝛾𝑐,𝜆 = 𝛾†𝑐†,𝜆
.
Proposition 4.3.3.6. Any distribution 𝐾 on hR with values in EndC(𝜆) satisfying the
conditions (i), (ii), and (iii) appearing in Proposition 4.3.3.4 satisfies the following
properties:
(i) The restriction 𝐾|hR,𝑟𝑒𝑔 of 𝐾 to the real regular locus
hR,𝑟𝑒𝑔 := 𝑥 ∈ hR : 𝛼𝑠(𝑥) = 0 for all 𝑠 ∈ 𝑆
satisfies the 2-sided KZ-type system of differential equations:
𝜕𝑦𝐾 +∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(𝑠𝐾 +𝐾𝑠) = 0 for all 𝑦 ∈ hR (4.3.2)
(ii) 𝐾 is a homogeneous distribution of degree −2𝜒𝜆(∑
𝑠∈𝑆 𝑐𝑠𝑠)/ dim𝜆, where 𝜒𝜆
149
is the character of 𝜆. In particular, 𝐾 is tempered.
Proof. To prove (i), take an arbitrary test function 𝜙 ∈ 𝐶∞𝑐 (hR,𝑟𝑒𝑔) ⊗ 𝜆. For any
𝑦 ∈ hR, we have (𝐾 ∘ 𝐷𝑦)(𝜙) = 0. From the definition of 𝐷𝑦 and the fact that
(1 ⊗ 𝑠)(𝜙)/𝛼𝑠 and (𝑠 ⊗ 𝑠)(𝜙)/𝛼𝑠 are also well-defined test functions in 𝐶∞𝑐 (hR) ⊗ 𝜆
for any 𝑠 ∈ 𝑆, we have:
0 =
∫hR
(𝐾 ∘𝐷𝑦)(𝜙)𝑑𝑥
=
∫hR
𝐾𝑐(𝑥)
(𝜕𝑦𝜙(𝑥) −
∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(𝑥)(𝑠𝜙(𝑥) − 𝑠𝜙(𝑠𝑥))
)𝑑𝑥
=
∫hR
(−𝜕𝑦𝐾(𝑥) −
∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(𝑥)𝐾(𝑥)𝑠
)𝜙(𝑥)𝑑𝑥
+∑𝑠∈𝑆
𝑐𝑠
∫hR
𝛼𝑠(𝑦)
𝛼𝑠(𝑥)𝐾(𝑥)𝑠𝜙(𝑠𝑥)𝑑𝑥
=
∫hR
(−𝜕𝑦𝐾(𝑥) −
∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(𝑥)𝐾(𝑥)𝑠
)𝜙(𝑥)𝑑𝑥
+∑𝑠∈𝑆
𝑐𝑠
∫hR
𝛼𝑠(𝑦)
𝛼𝑠(𝑠𝑥)𝐾(𝑠𝑥)𝑠𝜙(𝑥)𝑑𝑥
=
∫hR
(−𝜕𝑦𝐾(𝑥) −
∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(𝑥)𝐾(𝑥)𝑠
)𝜙(𝑥)𝑑𝑥
−∑𝑠∈𝑆
𝑐𝑠
∫hR
𝛼𝑠(𝑦)
𝛼𝑠(𝑥)𝑠𝐾(𝑥)𝜙(𝑥)𝑑𝑥
= −∫hR
(𝜕𝑦𝐾 +
∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(𝑥)(𝑠𝐾 +𝐾𝑠)
)𝜙(𝑥)𝑑𝑥,
where in the fourth equality we change variables 𝑥 ↦→ 𝑠𝑥 and in the fifth equality we
use the 𝑊 -invariance of 𝐾 and the equality 𝛼𝑠(𝑠𝑥) = −𝛼𝑠(𝑥). As 𝜙 ∈ 𝐶∞𝑐 (hR,𝑟𝑒𝑔)⊗𝜆
was arbitrary, this proves (i).
Next, note that as 𝑥1, ..., 𝑥𝑙 ∈ h*R and 𝑦1, ..., 𝑦𝑙 ∈ hR are dual bases we have
𝑙∑𝑖=1
𝑓(𝑦𝑖)𝑤(𝑥𝑖) = 𝑤(𝑓)
150
for all 𝑓 ∈ h* and 𝑠 ∈ 𝑆. It follows that we have, as operators on C[h] ⊗ 𝜆,
𝑙∑𝑖=1
𝛼𝑠(𝑦𝑖)
𝛼𝑠(1 − 𝑠)(𝑥𝑖) =
1
𝛼𝑠
(𝑙∑
𝑖=1
𝛼𝑠(𝑦𝑖)(𝑥𝑖 − 𝑠(𝑥𝑖)𝑠)
)=
1
𝛼𝑠(𝛼𝑠 − 𝑠(𝛼𝑠)𝑠) = 1 + 𝑠
(4.3.3)
Using the fact that 𝐾 ∘ 𝐷𝑦 = 0 for all 𝑦 ∈ hR, the action of the Euler operator∑𝑙𝑖=1 𝑥𝑖𝜕𝑦𝑖 on the distribution 𝐾(𝑥) is therefore given by
(𝑙∑
𝑖=1
𝑥𝑖𝜕𝑦𝑖
)𝐾(𝑥) = −𝐾(𝑥)
𝑙∑𝑖=1
𝜕𝑦𝑖𝑥𝑖
= −𝐾(𝑥)𝑙∑
𝑖=1
(∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦𝑖)
𝛼𝑠(1 − 𝑠) ⊗ 𝑠
)𝑥𝑖
= −𝐾(𝑥)
(∑𝑠∈𝑆
𝑐𝑠
(𝑙∑
𝑖=1
𝛼𝑠(𝑦𝑖)
𝛼𝑠(1 − 𝑠)𝑥𝑖
)⊗ 𝑠
)
= −𝐾(𝑥)∑𝑠∈𝑆
𝑐𝑠(1 + 𝑠) ⊗ 𝑠
= −∑𝑠∈𝑆
𝑐𝑠(𝑠𝐾(𝑥) +𝐾(𝑥)𝑠)
=−2𝜒𝜆(
∑𝑠∈𝑆 𝑐𝑠𝑠)
dim𝜆𝐾(𝑥),
where the second equality follows from the equation 𝐾 ∘𝐷𝑦𝑖 = 0, the fourth equality
uses equation (4.3.3), the fifth equality uses the 𝑊 -invariance of 𝐾(𝑥), and the final
equality uses the fact that∑
𝑠∈𝑆 𝑐𝑠𝑠 is central in C𝑊 and therefore acts on 𝜆 by the
scalar 𝜒𝜆(∑
𝑠∈𝑆 𝑐𝑠𝑠)/ dim𝜆, giving (ii).
Proposition 4.3.3.7. For any 𝑐 ∈ p, the support of the distribution 𝐾𝑐,𝜆 coincides
with the set of real points of the support of 𝐿𝑐(𝜆):
Supp(𝐾𝑐,𝜆) = Supp(𝐿𝑐(𝜆))R.
Proof. This was shown by Etingof [28, Proposition 3.10] in the case 𝜆 = triv, and the
proof generalizes to arbitrary 𝜆 without modification.
151
4.3.4 Existence of Dunkl Weight Function for Small 𝑐
Let 𝛼1, ..., 𝛼𝑙 be the simple roots associated to the system of positive roots 𝛼𝑠𝑠∈𝑆,
let
𝒞 := 𝑥 ∈ hR : 𝛼𝑖(𝑥) > 0 for 𝑖 = 1, ..., 𝑙
be the open associated fundamental Weyl chamber, and let 𝒞 denote its closure.
Recall that
𝛿 :=∏𝑠∈𝑆
𝛼𝑠 ∈ C[h]
is the discriminant element defining the reflection hyperplanes of 𝑊 .
Lemma 4.3.4.1. Let 𝑐 ∈ p and let 𝐾 be a End(𝜆)-valued analytic function on hR,𝑟𝑒𝑔
satisfying the differential equation
𝜕𝑦𝐾 +∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(𝑠𝐾 +𝐾𝑠) = 0
for each 𝑦 ∈ hR. Then
(i) 𝐾 is homogeneous of degree −2𝜒𝜆(∑
𝑠∈𝑆 𝑐𝑠𝑠)/ dim𝜆;
(ii) there exists an integer 𝑁 ≥ 0 such that 𝛿𝑁𝐾, extended to hR with value 0
along the reflection hyperplanes ker(𝛼𝑠), is a continuous function on hR;
(iii) if |𝑐𝑠| is sufficiently small for all 𝑠 ∈ 𝑆 then 𝐾(𝑥) is a locally 𝐿1 function on
hR.
Proof. Statement (i) follows immediately from the differential equation and the fact
that the central element∑
𝑠∈𝑆 𝑐𝑠𝑠 ∈ C𝑊 acts on 𝜆 by the scalar 𝜒𝜆(∑
𝑠∈𝑆 𝑐𝑠𝑠)/ dim𝜆.
To establish (ii) and (iii) it suffices to show that 𝐾 is locally integrable on 𝒞 when
|𝑐𝑠| is sufficiently small for all 𝑠 ∈ 𝑆 and that there exists 𝑁 ≥ 0 such that 𝛿𝑁𝐾|𝒞extends to a continuous function on 𝒞 vanishing on the boundary. Let 𝜔∨
1 , ..., 𝜔∨𝑙 ∈ 𝒞
be the fundamental dominant coweights, so that 𝛼𝑖(𝜔∨𝑗 ) = 𝛿𝑖𝑗, and let 𝑅 be the region
𝑅 := 𝑥 ∈ hR,𝑟𝑒𝑔 : 𝛼𝑖(𝑥) ∈ (0, 1), 𝑖 = 1, ..., 𝑙 ⊂ 𝒞.
152
Given a point 𝑣 ∈ 𝒞, we have, for those 𝑡 > 0 such that 𝛼𝑖(𝑣 − 𝑡𝜔∨𝑖 ) > 0,
𝑑
𝑑𝑡||𝐾(𝑣 − 𝑡𝜔∨
𝑖 )|| ≤ || 𝑑𝑑𝑡𝐾(𝑣 − 𝑡𝜔∨
𝑖 )||
= ||∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝜔∨𝑖 )
𝛼𝑠(𝑣 − 𝑡𝜔∨𝑖 )
(𝑠𝐾(𝑣 − 𝑡𝜔∨𝑖 ) +𝐾(𝑣 − 𝑡𝜔∨
𝑖 )𝑠)||
≤∑𝑠∈𝑆
2|𝑐𝑠|𝛼𝑠(𝜔∨𝑖 )
𝛼𝑠(𝑣 − 𝑡𝜔∨𝑖 )
||𝐾(𝑣 − 𝑡𝜔∨𝑖 )||,
where the norm || · || on EndC(𝜆) is the norm arising from the natural inner product
on EndC(𝜆) associated to the inner product (·, ·)𝜆 on 𝜆. Note that this norm is
differentiable away from 0 ∈ EndC(𝜆), the system of differential equations 𝐾 satisfies
implies that if 𝐾(𝑥) = 0 for any 𝑥 ∈ 𝒞 then 𝐾 is identically 0 on 𝒞, and any 𝑠 ∈ 𝑆
acts by an isometry with respect to this norm. It follows that
||𝐾(𝑣 − 𝑡𝜔∨𝑖 )|| ≤ ||𝐾(𝑣)||
∏𝑠∈𝑆
(𝛼𝑠(𝑣)
𝛼𝑠(𝑣 − 𝑡𝜔∨𝑖 )
)2|𝑐𝑠|
.
With 𝜌∨ :=∑
𝑖 𝜔∨𝑖 ∈ 𝒞, we therefore have
||𝐾(𝑥)|| ≤ ||𝐾(𝜌∨)||∏𝑠∈𝑆
(𝛼𝑠(𝜌∨)/𝛼𝑠(𝑥))2|𝑐𝑠| (4.3.4)
for all 𝑥 ∈ 𝑅, and the claim follows immediately from this estimate and the homo-
geneity of 𝐾.
We need to impose further conditions, in addition to 𝑊 -equivariance, on such
a locally integrable function 𝐾 in order for it to represent the form 𝛾𝑐,𝜆. In [20,
Section 5] and [19, Equation 9], Dunkl derived certain conditions on such 𝐾 near
the boundaries of the Weyl chambers. We will now establish similar conditions in
terms of the nature of the singularities of 𝐾 along the reflection hyperplanes, and
we will relate these conditions to the invariance of sesquilinear pairings on certain
representations of the Hecke algebra.
153
Definition 4.3.4.2. For each simple reflection 𝑠𝑖, let
𝒞𝑖 := 𝑥 ∈ hR : 𝛼𝑖(𝑥) = 0, 𝛼𝑗(𝑥) > 0 for 𝑗 = 𝑖
be the open codimension-1 face of 𝒞 determined by 𝛼𝑖.
Let 𝜆 ∈ Irr(𝑊 ) and let 𝑐 ∈ p be such that 𝑐𝑠 /∈ 12Z∖0 for all 𝑠 ∈ 𝑆. Fix a
simple reflection 𝑠𝑖, and choose coordinates 𝑧1, ..., 𝑧𝑙 ∈ h*R ⊂ h* for h with 𝑧1 = 𝛼𝑖 and
𝑧𝑗(𝛼∨𝑖 ) = 0 for 𝑗 > 1. Consider the modified 𝐾𝑍 connection
∇′𝐾𝑍 := 𝑑−
∑𝑠∈𝑆
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
𝑠
on the trivial vector bundle on h𝑟𝑒𝑔 with fiber 𝜆. The connection ∇′𝐾𝑍 is flat and
therefore has a unique local extension of any initial value to a local homomorphic
flat section. Furthermore, considering the restriction of ∇′𝐾𝑍 in the 𝑧1 direction
near a point 𝑥0 ∈ 𝒞𝑖 lying on the reflection hyperplane ker(𝛼𝑖), it follows from the
standard theory of ordinary differential equations with regular singularities (see, e.g.
[72, Theorem 5.5]) that there is a holomorphic function 𝑃𝑖(𝑧), defined in a complex
analytic 𝑧1-disc about 𝑥0 and satisfying 𝑃𝑖(𝑥0) = Id, such that the function
𝑧 ↦→ 𝑃𝑖(𝑧)𝑧𝑐𝑖𝑠𝑖1
gives a fundamental EndC(𝜆)-valued (multivalued) solution to the restriction of the
system ∇′𝐾𝑍 to a small punctured 𝑧1-disc about 𝑥0. We may extend 𝑃𝑖(𝑧) in the
𝑧2, ..., 𝑧𝑙 directions by taking solutions of the restricted connection
∇′′𝐾𝑍 := 𝑑−
∑𝑠∈𝑆∖𝑠𝑖
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
𝑠
along affine hyperplanes parallel to ker(𝛼𝑖) to produce a single valued function 𝑃𝑖(𝑧),
holomorphic in 𝑧2, ..., 𝑧𝑙, defined in a simply connected complex analytic 𝑠𝑖-stable
neighborhood of𝐷 of 𝒞∪𝒞𝑖∪𝑠𝑖(𝒞) (note that the restricted connection above is regular
along 𝒞𝑖 itself). The independence of the function 𝑧𝑐𝑖𝑠𝑖1 on the variables 𝑧2, ..., 𝑧𝑙 and
154
the uniqueness of extensions of solutions of ∇′𝐾𝑍 then implies that the such that the
function
𝑁𝑖(𝑧) := 𝑃𝑖(𝑧)𝑧𝑐𝑖𝑠𝑖1
gives a fundamental EndC(𝜆)-valued (multivalued) solution to the system ∇′𝐾𝑍 on
the region 𝐷𝑟𝑒𝑔 := 𝐷 ∩ h𝑟𝑒𝑔. In particular, 𝑃𝑖(𝑧) is holomorphic on 𝐷𝑟𝑒𝑔. Continuous
dependence of solutions to ordinary differential equations on parameters and initial
conditions implies that 𝑃𝑖(𝑧) is continuous on all of 𝐷. In particular, viewing 𝑃𝑖(𝑧)
as a function on 𝐷𝑟𝑒𝑔 holomorphic in 𝑧1, the singularities of 𝑃𝑖(𝑧) along ker(𝛼𝑖) are
removable. It follows that 𝑃𝑖(𝑧) is holomorphic in 𝑧1 on all of 𝐷. As 𝑃𝑖(𝑧) is holo-
morphic in each 𝑧𝑗 for 𝑗 > 1, it follows by Hartogs’ theorem that 𝑃𝑖(𝑧) is holomorphic
on all of 𝐷.
We next derive further properties of 𝑃𝑖(𝑧) arising from 𝑊 -equivariance of the
connection ∇′𝐾𝑍 and the initial condition 𝑃𝑖(𝑥0) = Id. In particular, we see that
𝑠𝑖𝑁𝑖(𝑠𝑖𝑧)𝑠𝑖 is another multivalued fundamental solution of ∇′𝐾𝑍 on 𝐷𝑟𝑒𝑔. As
𝑠𝑖𝑁𝑖(𝑠𝑖𝑧)𝑠𝑖 = 𝑠𝑖𝑃𝑖(𝑠𝑖𝑧)𝑠𝑖(−𝑧1)𝑐𝑖𝑠𝑖 = 𝑠𝑖𝑃𝑖(𝑠𝑖𝑧)𝑠𝑖𝑧𝑐𝑖𝑠𝑖1 𝑒𝜋
√−1𝑐𝑖𝑠𝑖 ,
it follows that 𝑠𝑖𝑃𝑖(𝑠𝑖𝑧)𝑠𝑖𝑧𝑐𝑖𝑠𝑖1 is such a solution as well (to reduce ambiguity, here
we use the notation√−1 for the imaginary unit 𝑖 ∈ C to avoid confusion with the
index of the simple reflection 𝑠𝑖). By the proof of [72, Theorem 5.5], in particular its
use of [72, Theorem 4.1], and the assumption that 𝑐𝑖 /∈ 12Z∖0, any function 𝑃𝑖(𝑧)
holomorphic in a small complex analytic 𝑧1-disc about 𝑥0 such that 𝑃𝑖(𝑥0) = Id and
the function 𝑃𝑖(𝑧)𝑧−𝑐𝑖𝑠𝑖1 is a solution to the system ∇′𝐾𝑍 on the punctured 𝑧1-disc
must coincide with 𝑃𝑖(𝑧) along the entire disc. It follows that 𝑠𝑖𝑃𝑖(𝑠𝑖𝑧)𝑠𝑖 = 𝑃𝑖(𝑧) for
𝑧 in a small complex analytic 𝑧1-disc about 𝑥0 and therefore for all 𝑧 ∈ 𝐷.
We summarize the conclusions of the above discussion in the following lemma:
Lemma 4.3.4.3. Let 𝜆 ∈ Irr(𝑊 ), let 𝑐 ∈ p be such that 𝑐𝑖 /∈ 12Z∖0 for a given sim-
ple reflection 𝑠𝑖 ∈ 𝑆, and let 𝑧1, ..., 𝑧𝑙 be coordinates for h as in the discussion above.
There is a 𝑠𝑖-stable simply connected complex analytic neighborhood 𝐷, independent
of 𝑐, of 𝒞 ∪𝒞𝑖 ∪ 𝑠𝑖(𝒞) in h and a holomorphic 𝐺𝐿(𝜆)-valued function 𝑃𝑖(𝑧) on 𝐷 such
155
that
(1) 𝑃𝑖(𝑧)𝛼𝑖(𝑧)𝑐𝑖𝑠𝑖 is a EndC(𝜆)-valued multivalued holomorphic fundamental solu-
tion to the modified 𝐾𝑍 system
∇′𝐾𝑍 := 𝑑−
∑𝑠∈𝑆
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
𝑠
on the domain 𝐷𝑟𝑒𝑔 := 𝐷 ∩ h𝑟𝑒𝑔
(2) 𝑠𝑖𝑃𝑖(𝑠𝑖𝑧)𝑠𝑖 = 𝑃𝑖(𝑧) for all 𝑧 ∈ 𝐷. In particular, 𝑃𝑖 takes values in End𝑠𝑖(𝜆)
along ker(𝛼𝑖) ∩𝐷.
(3) 𝑃𝑖(𝑧) is a solution of the system
∇′′𝐾𝑍 := 𝑑−
∑𝑠∈𝑆∖𝑠𝑖
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
𝑠
along ker(𝛼𝑖) ∩𝐷.
Furthermore, any such 𝑃𝑖 is determined by its value at any point 𝑥0 ∈ 𝒞𝑖, and for
any constant invertible 𝐴 ∈ Aut𝑠𝑖(𝜆) the function 𝑃𝑖(𝑧)𝐴 is another such function.
In particular, any two such functions 𝑃𝑖(𝑧), 𝑃𝑖(𝑧) are related by an equality 𝑃𝑖(𝑧) =
𝑃𝑖(𝑧)𝐴 for a unique such 𝐴.
Now, let 𝐾 : 𝒞 → EndC(𝜆) be a real analytic function satisfying differential
equation (4.3.2) appearing in Proposition 4.3.3.6
𝜕𝑦𝐾 +∑𝑠∈𝑆
𝑐𝑠𝛼𝑠(𝑦)
𝛼𝑠(𝑠𝐾 +𝐾𝑠) = 0 for all 𝑦 ∈ hR.
Choose a simple reflection 𝑠𝑖, and fix functions 𝑃𝑖,𝑐(𝑧) and 𝑃𝑖,𝑐† as in Lemma 4.3.4.3
for parameters 𝑐 and 𝑐†, respectively. By the uniqueness of solutions of differential
equation (4.3.2) given initial conditions, there is a matrix 𝐾𝑖 ∈ EndC(𝜆), uniquely
determined given 𝑃𝑖(𝑧), such that 𝐾 is given by
𝑧 ↦→ 𝑃𝑖,𝑐†(𝑧)†,−1𝛼𝑖(𝑧)−𝑐𝑖𝑠𝑖𝐾𝑖𝛼𝑖(𝑧)−𝑐𝑖𝑠𝑖𝑃𝑖,𝑐(𝑧)−1
156
for 𝑧 ∈ 𝒞. As 𝑃𝑖,𝑐(𝑧) and 𝑃𝑖,𝑐†(𝑧) are determined only up to right multiplication by
some 𝐴 ∈ Aut𝑠𝑖(𝜆), here 𝐾𝑖 is determined by 𝐾 only up to the action of Aut𝑠𝑖(𝜆) on
EndC(𝜆) by 𝐴.𝑀 = 𝐴𝑀𝐴†. As 𝑠†𝑖 = 𝑠𝑖, 𝑠𝑖-invariance of 𝐴 implies that of 𝐴†, and we
see that 𝑠𝑖-invariance of 𝐾𝑖 is a property of 𝐾, not depending on the choice of 𝑃𝑖,𝑐(𝑧)
and 𝑃𝑖,𝑐†(𝑧). In particular, we make the following definition:
Definition 4.3.4.4. In the setting of the previous paragraph, we say that the function
𝐾 is asymptotically 𝑊 -invariant if 𝐾𝑖 ∈ End𝑠𝑖(𝜆) for all simple reflections 𝑠𝑖.
We will need the following lemma:
Lemma 4.3.4.5. Fix 𝑥0 ∈ hR,𝑟𝑒𝑔. There is a holomorphic function 𝐵 : p → EndC(𝜆)
such that 𝐵(0) = Id and, for all 𝑐 ∈ p, 𝐵(𝑐) defines a 𝐵𝑊 -invariant sesquilinear
pairing
𝐾𝑍𝑥0(∆𝑐(𝜆)) ×𝐾𝑍𝑥0(∆𝑐†(𝜆)) → C
(𝑣1, 𝑣2) ↦→ 𝑣†2𝐵(𝑐)𝑣1,
where we make the standard identification of 𝐾𝑍𝑥0(∆𝑐(𝜆)) and 𝐾𝑍𝑥0(∆𝑐†(𝜆)) with 𝜆
as vector spaces. Any two such functions 𝐵(𝑐) and 𝐵(𝑐) are related by 𝐵(𝑐) = 𝑏(𝑐)𝐵(𝑐)
for a meromorphic function 𝑏(𝑐). Replacing 𝐵(𝑐) by (𝐵(𝑐)+𝐵(𝑐†)†)/2, we may assume
𝐵(𝑐) takes values in Hermitian forms for all 𝑐 ∈ pR.
Proof. An operator 𝐴𝑐 ∈ EndC(𝜆) defines a 𝐵𝑊 -invariant sesquilinear pairing
𝐾𝑍𝑥0(∆𝑐(𝜆)) ×𝐾𝑍𝑥0(∆𝑐†(𝜆)) → C
as in the lemma statement if and only if
𝑇 †,−1𝑖,𝑐†
𝐴𝑐𝑇−1𝑖,𝑐 = 𝐴𝑐, (4.3.5)
where 𝑇𝑖,𝑐 and 𝑇𝑖,𝑐† denote the action of the generator 𝑇𝑖 ∈ 𝐵𝑊 in 𝐾𝑍𝑥0(∆𝑐(𝜆)) and
𝐾𝑍𝑥0(∆𝑐†(𝜆)), respectively, for all generators 𝑇𝑖 of 𝐵𝑊 . Similarly to Lemma 4.2.2.5,
157
such a sequilinear pairing is equivalent to a homomorphism of 𝐵𝑊 -representations
𝐾𝑍𝑥0(∆𝑐†(𝜆)) → 𝐾𝑍𝑥0(∆𝑐(𝜆))ℎ (4.3.6)
where 𝐾𝑍𝑥0(∆𝑐(𝜆))ℎ denotes the Hermitian dual of 𝐾𝑍𝑥0(∆𝑐(𝜆)) as a representation
of 𝐵𝑊 , i.e. the representation of 𝐵𝑊 which is 𝐾𝑍𝑥0(∆𝑐(𝜆))* as a C-vector space and
in which the action of 𝑔 ∈ 𝐵𝑊 is by 𝑔.𝑓 = 𝑓 ∘ 𝑔−1. When |𝑐𝑠| is small for all 𝑠 ∈ 𝑆,
the Hecke algebra H𝑞(𝑊 ), 𝑞(𝑠) = 𝑒−2𝜋𝑖𝑐(𝑠), is semisimple and isomorphic to C𝑊 , and
the KZ functor is an equivalence of categories 𝒪𝑐(𝑊, h) ∼= H𝑞(𝑊 )-mod𝑓.𝑑.. For such
𝑐, the irreducible representations of H𝑞(𝑊 ) therefore are precisely the images of the
standard modules under the KZ functor, and in particular they can be distinguished
by their character as 𝐵𝑊 -representations, as this is so for 𝑐 = 0 and these characters
are continuous (in fact, holomorphic) functions of 𝑐. In particular, for such 𝑐, as
𝐾𝑍𝑥0(∆𝑐†(𝜆)) and 𝐾𝑍𝑥0(∆𝑐(𝜆))ℎ are irreducible, isomorphic for 𝑐 = 0, and have
characters as 𝐵𝑊 -representations that are continuous functions of 𝑐, it follows that
they are isomorphic. It follows that, when |𝑐𝑠| is small for all 𝑠 ∈ 𝑆, there is a
1-dimensional space of 𝐵𝑊 -isomorphisms as in (4.3.6), and hence a unique solution
𝐴𝑐 ∈ EndC(𝜆) to the system of equations
⎧⎪⎨⎪⎩𝑇†,−1𝑖,𝑐†
𝐴𝑐𝑇−1𝑖,𝑐 = 𝐴𝑐, for all generators 𝑇𝑖 ∈ 𝐵𝑊
Tr(𝐴𝑐) = dim(𝜆).
(4.3.7)
The operators 𝑇−1𝑖,𝑐 and 𝑇 †,−1
𝑖,𝑐†are holomorphic in 𝑐 ∈ p, so we may view system
(4.3.7) as a system of linear system of equations in the variable 𝐴𝑐 ∈ EndC(𝜆) with
coefficients that are holomorphic in 𝑐 ∈ p. It follows that there is a solution 𝐴𝑐 that
is meromorphic in 𝑐 ∈ p with singularities where the system is degenerate. Clearing
denominators by multiplying by an appropriate determinant 𝑑(𝑐), holomorphic in 𝑐,
the resulting holomorphic function 𝑑(𝑐)𝐴𝑐, up to scaling by a global constant, satisfies
the properties of 𝐵(𝑐) stated in the lemma. The uniqueness statement follows as well.
Finally, if 𝐵(𝑐) satisfies the properties in the Lemma then 𝐵(𝑐) := (𝐵(𝑐) +𝐵(𝑐†)†)/2
158
is also holomorphic in 𝑐 and satisfies equation (4.3.5) for all 𝑐 ∈ pR, and therefore for
all 𝑐 ∈ p as well by analyticity.
Remark 4.3.4.6. Lemma 4.3.4.5 holds for complex reflection groups as well, with
the same proof, although this level of generality will not be needed.
Theorem 4.3.4.7 (Existence of Dunkl Weight Function for Small 𝑐). Let 𝜆 ∈ Irr(𝑊 )
and let 𝑐 ∈ p with |𝑐𝑠| sufficiently small for all 𝑠 ∈ 𝑆. Let 𝐾 : hR,𝑟𝑒𝑔 → EndC(𝜆) be
a nonzero 𝑊 -equivariant function satisfying system (4.3.2). Then the following are
equivalent:
(a) 𝐾 represents the Gaussian pairing 𝛾𝑐,𝜆 up to rescaling by a nonzero complex
number.
(b) For any point 𝑥 ∈ hR,𝑟𝑒𝑔, 𝐾(𝑥) determines a 𝐵𝑊 -invariant sesquilinear pairing
𝐾𝑍𝑥(∆𝑐(𝜆)) ×𝐾𝑍𝑥(∆𝑐†(𝜆)) → C.
(c) 𝐾|𝒞 is asymptotically 𝑊 -invariant in the sense of Definition 4.3.4.4.
Furthermore, the space of such 𝐾 satisfying (a)-(c) forms a one-dimensional com-
plex vector space.
Proof. We will first show that (a) and (c) are equivalent. As the function 𝐾 is
locally integrable and homogeneous by Lemma 4.3.4.1 when |𝑐𝑠| is small for all 𝑠 ∈ 𝑆,
it determines a tempered distribution with values in EndC(𝜆). Therefore, we may
define a sesquilinear pairing 𝛾𝐾 : ∆𝑐(𝜆) × ∆𝑐†(𝜆) → C by the formula
𝛾𝐾(𝑃,𝑄) =
∫hR
𝑄(𝑥)†𝐾(𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥
for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆. As 𝐾 is nonzero and 𝑊 -equivariant, the same is true for 𝛾𝐾
by the density of C[h]𝑒−|𝑥|2/2 ⊂ S (hR). By Proposition 4.3.3.4, 𝛾𝐾 is proportional to
𝛾𝑐,𝜆 if and only if ∫hR
𝐾(𝑥)𝐷𝑦𝜙(𝑥)𝑑𝑥 = 0
159
for all 𝜙 ∈ S (hR) and 𝑦 ∈ hR. Following Dunkl, for any 𝜖 > 0 define the region
Ω𝜖 := 𝑥 ∈ hR : |𝛼𝑠(𝑥)| > 𝜖 for all 𝑠 ∈ 𝑆 ⊂ hR,𝑟𝑒𝑔.
For any 𝜙 ∈ S (hR) ⊗ 𝜆 the function 𝐾(𝐷𝑦𝜙) is integrable, so we have
lim𝜖→0
∫Ω𝑒
𝐾(𝑥)𝐷𝑦𝜙(𝑥)𝑑𝑥 =
∫hR
𝐾(𝑥)𝐷𝑦𝜙(𝑥)𝑑𝑥
by the dominated convergence theorem (see, e.g., [64, Section 4.4]). The region
Ω𝜖 avoids the singularities of 1/𝛼𝑠 for all 𝑠 ∈ 𝑆, and so, following the calculations
appearing after Equation (14) in [20, Section 5], a direct calculation using the 𝑊 -
invariance of 𝐾, the 𝑊 -invariance of Ω𝜖, and the system of differential equations that
𝐾 satisfies shows that
∫Ω𝜖
𝐾(𝑥)𝐷𝑦𝜙(𝑥)𝑑𝑥 =
∫Ω𝜖
𝜕𝑦(𝐾(𝑥)𝜙(𝑥))𝑑𝑥.
By Proposition 4.3.3.4, we see that 𝛾𝐾 is proportional to 𝛾𝑐,𝜆 if and only if
lim𝜖→0
∫Ω𝜖
𝜕𝑦(𝐾(𝑥)𝜙(𝑥))𝑑𝑥 = 0 (4.3.8)
for a spanning set of 𝑦 ∈ hR and dense set of 𝜙 ∈ S (hR).
Fix a simple reflection 𝑠𝑖. Note that the boundary wall of the cone Ω𝜖 ∩𝒞 parallel
to 𝒞𝑖 is the translate of 𝒞𝑖 by 𝜖𝜌∨ and that the boundary wall of the cone Ω𝜖 ∩ 𝑠𝑖𝒞
parallel to 𝒞𝑖 is the translate of 𝒞𝑖 by 𝜖𝑠𝑖𝜌∨ = 𝜖(𝜌∨ − 𝛼∨
𝑖 ). Computing the integral
above as an iterated integral with inner integrals chosen as in [20, Section 5] and
choosing test functions 𝜙 to be bump functions supported near a point 𝑥0 ∈ 𝒞𝑖 and
𝑠𝑖-invariant, we see that the vanishing of the limit
lim𝜖→0
∫Ω𝜖
𝜕𝛼∨𝑖(𝐾𝜙)𝑑𝑥
160
for all 𝜙 implies that
lim𝜖→0
∫𝐶𝑖
(𝐾 − 𝑠𝑖𝐾𝑠𝑖)(𝑥+ 𝜖𝜌∨)𝜓(𝑥)𝑑𝑥 = lim𝜖→0
∫𝐶𝑖
(𝐾(𝑥+ 𝜖𝜌∨)−𝐾(𝑥+ 𝜖𝑠𝑖𝜌∨))𝜓(𝑥)𝑑𝑥 = 0
for all 𝜓 ∈ 𝐶∞𝑐 (𝒞𝑖), where the first equality follows from the 𝑊 -invariance of 𝐾.
Choosing coordinates 𝑧1, ..., 𝑧𝑙 adapted to the wall 𝒞𝑖 as in the discussion preceding
Definition 4.3.4.4, we have that 𝐾 is given on 𝒞 by the formula
𝑧 ↦→ 𝑃𝑖,𝑐†(𝑧)†,−1𝛼𝑖(𝑧)−𝑐𝑖𝑠𝑖𝐾𝑖𝛼𝑖(𝑧)−𝑐𝑖𝑠𝑖𝑃𝑖,𝑐(𝑧)−1
with 𝑃𝑖,𝑐(𝑧), 𝑃𝑖,𝑐† , and 𝐾𝑖 as above. As 𝑃𝑖,𝑐 and 𝑃𝑖,𝑐† take values in Aut𝑠𝑖(𝜆) on 𝒞𝑖, it
follows that the commutators [𝑠𝑖, 𝑃𝑖,𝑐†(𝑧)†,−1] and [𝑃𝑖,𝑐(𝑧)−1, 𝑠𝑖] are antiholomorphic
and holomorphic, respectively, in 𝑧 and vanish along 𝒞𝑖. In particular, there is a
holomorphic function 𝑄𝑖,𝑐(𝑧) and an antiholomorphic function 𝑄𝑖,𝑐†(𝑧) on 𝐷 with
values in EndC(𝜆) such that [𝑠𝑖, 𝑃𝑖,𝑐†(𝑧)†,−1] = 𝑧1𝑄𝑖,𝑐†(𝑧) and [𝑃𝑖,𝑐(𝑧)−1, 𝑠𝑖] = 𝑧1𝑄𝑖,𝑐(𝑧).
For 𝜎1, 𝜎2 ∈ ±1 let 𝐾𝜎1,𝜎2𝑖 ∈ EndC(𝜆) be the projection of 𝐾𝑖 to the (𝜎1, 𝜎2)
simultaneous eigenspace of the commuting operators 𝜆𝑠𝑖 and 𝜌𝑠𝑖 on EndC(𝜆) of left and
right, respectively, multiplication by 𝑠𝑖. We have 𝐾𝑖 = 𝐾1,1𝑖 +𝐾1,−1
𝑖 +𝐾−1,1𝑖 +𝐾−1,−1
𝑖 ,
and a straightforward computation shows
(𝐾 − 𝑠𝑖𝐾𝑠𝑖)(𝑧) = 2𝑃𝑖,𝑐†(𝑧)†,−1(𝐾1,−1𝑖 +𝐾−1,1
𝑖 )𝑃𝑖,𝑐(𝑧)−1 + 𝑧1−2|𝑐𝑖|1 𝑅𝑖(𝑧) (4.3.9)
for 𝑧 ∈ 𝒞, where 𝑅𝑖(𝑧) ∈ EndC(𝜆) is analytic on 𝒞 and extends continuously to 𝒞 ∪𝒞𝑖.
It follows that we have
lim𝜖→0
∫𝒞𝑖
(𝐾 − 𝑠𝑖𝐾𝑠𝑖)(𝑥+ 𝜖𝜌∨)𝜓(𝑥)𝑑𝑥
=
∫𝒞𝑖
2𝑃𝑖,𝑐†(𝑥+ 𝜖𝜌∨)†,−1(𝐾1,−1𝑖 +𝐾−1,1
𝑖 )𝑃𝑖,𝑐(𝑥+ 𝜖𝜌∨)−1𝜓(𝑥)𝑑𝑥
for all 𝜓 ∈ 𝐶∞𝑐 (𝒞𝑖). For this integral to vanish for all such 𝜓, we must have that
𝐾1,−1𝑖 + 𝐾−1,1
𝑖 = 0, and hence that 𝐾𝑖 = 𝐾1,1𝑖 + 𝐾−1,−1
𝑖 commutes with 𝑠𝑖, so that
161
𝐾|𝒞 is asymptotically 𝑊 -invariant. In particular, we see that (a) implies (c).
Conversely, if (c) holds, equation (4.3.9) implies that 𝐾 − 𝑠𝑖𝐾𝑠𝑖 extends contin-
uously to all of 𝒞 ∪ 𝒞𝑖 ∪ 𝑠𝑖(𝒞) with value 0 along 𝒞𝑖. By the 𝑊 -equivariance of 𝐾
and the inner integrals used by Dunkl recalled in the previous paragraph, to show the
vanishing of limit (4.3.8), for 𝑦 ∈ hR,𝑟𝑒𝑔 and 𝜙 ∈ 𝐶∞𝑐 (hR), it suffices to show that the
limit
lim𝜖→0
∫𝒞𝑖
(𝐾𝜙)(𝑥+ 𝜖𝜌∨) − (𝐾𝜙)(𝑠𝑖(𝑥+ 𝜖𝜌∨))𝑑𝑥 (4.3.10)
vanishes. It suffices to treat the cases in which 𝜙 ∘ 𝑠𝑖 = ±𝜙. In the case 𝜙 ∘ 𝑠𝑖 = 𝜙,
as 𝐾 is 𝑊 -invariant the integrand of (4.3.10) is (𝐾 − 𝑠𝑖𝐾𝑠𝑖)(𝑥+ 𝜖𝜌∨)𝜙(𝑥+ 𝜖𝜌∨). In
the case 𝜙 ∘ 𝑠𝑖 = −𝜙, there exists a test function 𝜓 ∈ 𝐶∞𝑐 (hR) such that 𝜙 = 𝛼𝑖𝜓,
and the integrand of (4.3.10) takes the form (𝛼𝑖 · (𝐾 + 𝑠𝑖𝐾𝑠𝑖))(𝑥 + 𝜖𝜌∨)𝜓(𝑥 + 𝜖𝜌∨).
In particular, in either case the integrand is of the form 𝑓(𝑥+ 𝜖𝜌∨)𝜓(𝑥+ 𝜖𝜌∨) where
𝜓 ∈ 𝐶∞𝑐 (hR) is a test function and 𝑓 is either 𝐾− 𝑠𝑖𝐾𝑠𝑖 or 𝛼𝑖 · (𝐾+ 𝑠𝑖𝐾𝑠𝑖). In either
case, from estimate (4.3.4) in Lemma 4.3.4.1 and equation (4.3.9), there is a constant
𝜇 > 0, independent of 𝑐, such that
𝛿𝑖(𝑥)𝜇∑
𝑠∈𝑆 |𝑐𝑠|𝛼𝑖(𝑥)2|𝑐𝑖|−1𝑓(𝑥)
extends to a continuous function on 𝒞. In particular, let 𝑀 ≥ 0 be
𝑀 := max𝑥∈Supp(𝜓)∩𝒞
||𝛿𝑖(𝑥)𝜇∑
𝑠∈𝑆 |𝑐𝑠|𝛼𝑖(𝑥)2|𝑐𝑖|−1𝑓(𝑥)||
and let 𝑀 ′ := max𝑥∈hR |𝜓(𝑥)|. We then have
||𝑓(𝑥)|| ≤𝑀𝛿𝑖(𝑥)−𝜇∑
𝑠∈𝑆 |𝑐𝑠|𝛼𝑖(𝑥)1−2|𝑐𝑖|
for 𝑥 ∈ Supp(𝜓) ∩ 𝒞, and hence
||𝑓(𝑥+ 𝜖𝜌∨)𝜓(𝑥+ 𝜖𝜌∨)||
≤𝑀𝑀 ′𝛿𝑖(𝑥+ 𝜖𝜌∨)−𝜇∑
𝑠∈𝑆 |𝑐𝑠|𝛼𝑖(𝑥+ 𝜖𝜌∨)1−2|𝑐𝑖|
162
≤𝑀𝑀 ′𝛿𝑖(𝑥)−𝜇∑
𝑠∈𝑆 |𝑐𝑠|𝜖1−2|𝑐𝑖|
for all 𝑥 ∈ 𝒞𝑖 ∩ Supp(𝜓). Taking |𝑐𝑠| sufficiently small for all 𝑠 ∈ 𝑆 so that
𝛿𝑖(𝑥)−𝜇∑
𝑠∈𝑆 |𝑐𝑠| is locally integrable on 𝒞𝑖 and 1 − 2|𝑐𝑖| > 0, the vanishing of limit
(4.3.10) now follows from the dominated convergence theorem. This completes the
proof that (c) implies (a).
Now we will show the equivalence of (b) and (c). Fix a point 𝑥0 ∈ hR,𝑟𝑒𝑔 and
identify 𝐾(𝑥0) with a sesquilinear pairing 𝐾𝑍𝑥0(∆𝑐(𝜆)) ×𝐾𝑍𝑥0(∆𝑐†(𝜆)) → C. Here
we make the usual identification of𝐾𝑍𝑥0(∆𝑐(𝜆)) and𝐾𝑍𝑥0(∆𝑐†(𝜆)) with 𝜆 as C-vector
spaces, and 𝐾(𝑥0) ∈ EndC(𝜆) defines a sesquilinear form by (𝑣1, 𝑣2)𝐾(𝑥0) = 𝑣†2𝐾(𝑥0)𝑣1.
Without loss of generality, we take 𝑥 ∈ 𝒞. The braid group 𝐵𝑊 = 𝜋1(h𝑟𝑒𝑔/𝑊, 𝑥0)
is generated by the elements 𝑇𝑖 given by positively oriented half loops around the
hyperplanes ker(𝛼𝑖) for simple roots 𝛼𝑖. For each simple root 𝛼𝑖, let 𝑇𝑖,𝑐 and 𝑇𝑖,𝑐†
denote the action of 𝑇𝑖 on𝐾𝑍𝑥0(∆𝑐(𝜆)) and on𝐾𝑍𝑥0(∆𝑐†(𝜆)), respectively, so that the
action of 𝑇𝑖 on the space of sesquilinear pairings 𝐾𝑍𝑥0(∆𝑐(𝜆)) ×𝐾𝑍𝑥0(∆𝑐†(𝜆)) → C,
identified with EndC(𝜆), is given by
𝑋 ↦→ 𝑇 †,−1𝑖,𝑐†
𝑋𝑇−1𝑖,𝑐 . (4.3.11)
Consider the 𝑊 -equivariant EndC(𝜆)-valued multivalued function
𝐾(𝑧) := (𝑓𝑐†(𝑧)𝑀𝑐†(𝑧))†,−1𝐾(𝑥0)(𝑓𝑐(𝑧)𝑀𝑐(𝑧))−1
on h𝑟𝑒𝑔, where 𝑀𝑐(𝑧) is the monodromy of the 𝐾𝑍 connection
∇𝐾𝑍 := 𝑑+∑𝑠∈𝑆
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
(1 − 𝑠)
on the trivial bundle 𝜆 × h𝑟𝑒𝑔 → h𝑟𝑒𝑔 from 𝑥0 to 𝑧, 𝑓𝑐(𝑧) is the (scalar) monodromy
of the scalar-valued 𝑊 -equivariant flat connection
𝑑−∑𝑠∈𝑆
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
163
on the trivial bundle C×h𝑟𝑒𝑔 → h𝑟𝑒𝑔 from 𝑥0 to 𝑧, and𝑀𝑐† and 𝑓𝑐† are defined similarly
with the parameter 𝑐† in place of the parameter 𝑐. A straightforward computation
shows that 𝐾(𝑧) satisfies system (4.3.2) of differential equations on 𝒞, so it follows
from the equality 𝐾(𝑥0) = 𝐾(𝑥0) and the uniqueness of solutions to (4.3.2) given
initial conditions that 𝐾(𝑥) = 𝐾(𝑥) for all 𝑥 ∈ 𝒞.
View 𝐾(𝑧) as a multivalued section of the vector bundle
(EndC(𝜆) × h𝑟𝑒𝑔)/𝑊 → h𝑟𝑒𝑔/𝑊.
It follows by considering residues that the monodromy of 𝑓𝑐(𝑧) is inverse to the mon-
odromy of 𝑓𝑐†(𝑧), and in particular the function 𝑓𝑐(𝑧)𝑓𝑐†(𝑧) is single valued on h𝑟𝑒𝑔/𝑊 .
As the monodromy of 𝑀𝑐(𝑧) based at 𝑥0 is given, by definition, by the 𝐵𝑊 -action on
𝐾𝑍𝑥0(∆𝑐(𝜆)) and similarly for 𝑀𝑐†(𝑧), it follows that the monodromy in h𝑟𝑒𝑔/𝑊 of𝐾(𝑧) based at 𝑥0 is precisely given by the 𝐵𝑊 -action on sesquilinear pairings appear-
ing in (4.3.11) above. In particular, 𝐾(𝑥0) is 𝐵𝑊 -invariant as a sesquilinear pairing
𝐾𝑍𝑥0(∆𝑐(𝜆)) ×𝐾𝑍𝑥0(∆𝑐†(𝜆)) → C if and only if 𝐾(𝑧) is single-valued on h𝑟𝑒𝑔/𝑊 .
It remains to show that 𝐾(𝑧) is single-valued if and only if 𝐾(𝑥) is asymptotically
𝑊 -invariant in the sense of Definition 4.3.4.4. Fix a simple reflection 𝛼𝑖, and let
𝑃𝑖,𝑐(𝑧) and 𝑃𝑖,𝑐†(𝑧) be holomorphic functions on a domain 𝐷 as in Lemma 4.3.4.3.
It follows from that same lemma and the remarks following it that there is a unique
𝐾𝑖 ∈ EndC(𝜆) such that
𝐾(𝑧) = 𝑃𝑖,𝑐†(𝑧)†,−1𝛼𝑖(𝑧)−𝑐𝑖𝑠𝑖
𝐾𝑖𝛼𝑖(𝑧)−𝑐𝑖𝑠𝑖𝑃𝑖,𝑐(𝑧)−1
for all 𝑧 ∈ 𝐷 ∩ h𝑟𝑒𝑔. Decomposing 𝐾𝑖 into simultaneous left and right eigenspaces for
𝑠𝑖 as 𝐾𝑖 = 𝐾1,1𝑖 +𝐾1,−1
𝑖 +𝐾−1,1𝑖 +𝐾−1,−1
𝑖 as before, we have that 𝑃𝑖,𝑐†(𝑧)† 𝐾(𝑧)𝑃𝑖,𝑐(𝑧),
for 𝑧 ∈ 𝐷 ∩ h𝑟𝑒𝑔, is given by:
|𝛼𝑖(𝑧)|−2𝑐𝑖𝐾1,1𝑖 +
(𝛼𝑖(𝑧)
𝛼𝑖(𝑧)
)−𝑐𝑖
𝐾1,−1𝑖 +
(𝛼𝑖(𝑧)
𝛼𝑖(𝑧)
)𝑐𝑖
𝐾−1,1𝑖 + |𝛼𝑖(𝑧)|2𝑐𝑖𝐾−1,−1
𝑖 .
164
Clearly, this function, and hence 𝐾(𝑧) itself, is single-valued in h𝑟𝑒𝑔 near the hy-
perplane ker(𝛼𝑖) if and only if 𝐾1,−1𝑖 = 𝐾−1,1
𝑖 = 0, i.e. if 𝐾𝑖 = 𝐾1,1𝑖 + 𝐾−1,−1
𝑖 is
𝑠𝑖-invariant. In that case, by the 𝑠𝑖-equivariance of 𝑃𝑖,𝑐 and 𝑃𝑖,𝑐† , it then follows
that 𝐾(𝑧) is also single-valued in h𝑟𝑒𝑔/𝑊 near the hyperplane ker(𝛼𝑖). So, 𝐾(𝑧) is
invariant under the monodromy action of 𝑇𝑖 ∈ 𝐵𝑊 if and only if 𝐾𝑖 ∈ Aut𝑠𝑖(𝜆).
In particular, 𝐾(𝑧) is single-valued if and only if 𝐾 is asymptotically 𝑊 -invariant,
completing the proof of the equivalence of (b) and (c).
The final statement of the theorem follows from Lemma 4.3.4.5.
4.3.5 Analytic Continuation on the Regular Locus
Theorem 4.3.5.1 (Existence of Dunkl Weight Function on hR,𝑟𝑒𝑔). There exists a
unique family of analytic functions 𝐾𝑐 : hR,𝑟𝑒𝑔 → EndC(𝜆) with holomorphic depen-
dence on 𝑐 such that the following holds: for all 𝑀 > 0 there exists an integer 𝑁 ≥ 0,
which may be taken to be 0 for 𝑀 sufficiently small, such that for 𝑐 ∈ p with |𝑐𝑠| < 𝑀
for all 𝑠 ∈ 𝑆 the function 𝛿2𝑁𝐾𝑐 is locally integrable on hR and represents the Gaussian
pairing 𝛾𝑐 on 𝛿𝑁∆𝑐(𝜆) × 𝛿𝑁∆𝑐†(𝜆) in the following sense:
𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄) =
∫hR
𝑄(𝑥)†𝛿2𝑁(𝑥)𝐾𝑐(𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥 for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆.
For each 𝑐 ∈ p and 𝑥 ∈ hR,𝑟𝑒𝑔, 𝐾𝑐(𝑥) defines a 𝐵𝑊 -invariant sesquilinear pairing
𝐾𝑍𝑥(∆𝑐(𝜆)) ×𝐾𝑍𝑥(∆𝑐†(𝜆)) → C.
For 𝑐 ∈ pR, 𝐾𝑐(𝑥) defines a 𝐵𝑊 -invariant Hermitian form on 𝐾𝑍𝑥(∆𝑐(𝜆)).
Proof. Uniqueness of 𝐾𝑐 follows from the density of C[h]𝑒−|𝑥|2/2 in the Schwartz space
S (hR), so it suffices to establish existence of 𝐾𝑐. Let 𝐵 : p → EndC(𝜆) be a holomor-
phic function as in Lemma 4.3.4.5 defined relative to the point 𝑥0 ∈ 𝒞. It follows from
the holomorphic dependence on the parameter 𝑐 and on initial conditions of solutions
of the system of differential equations (4.3.2) appearing in Proposition 4.3.3.6 that
there is a unique function 𝐾 ′ : p× hR,𝑟𝑒𝑔 → EndC(𝜆), 𝐾 ′(𝑐, 𝑥) = 𝐾 ′𝑐(𝑥), holomorphic
165
in 𝑐 ∈ p and analytic in 𝑥 ∈ hR,𝑟𝑒𝑔, such that
(1) 𝐾 ′𝑐(𝑥0) = 𝐵(𝑐) for all 𝑐 ∈ p
(2) 𝑤𝐾 ′𝑐(𝑤
−1𝑥)𝑤−1 = 𝐾 ′𝑐(𝑥) for all 𝑐 ∈ p, 𝑥 ∈ hR,𝑟𝑒𝑔, and 𝑤 ∈ 𝑊
(3) For all fixed 𝑐 ∈ p, 𝐾 ′𝑐(𝑥), as a function of 𝑥 ∈ 𝒞 satisfies the system of
differential equations (4.3.2).
Furthermore, as 𝐵(𝑐) defines a 𝐵𝑊 -invariant sesquilinear pairing
𝐾𝑍𝑥0(∆𝑐(𝜆)) ×𝐾𝑍𝑥0(∆𝑐†(𝜆)) → C,
it follows from the proof of the equivalence of statements (b) and (c) in Theorem
4.3.4.7 that 𝐾 ′𝑐(𝑥) defines a 𝐵𝑊 -invariant sesquilinear pairing
𝐾𝑍𝑥(∆𝑐(𝜆)) ×𝐾𝑍𝑥(∆𝑐†(𝜆)) → C
for all 𝑐 ∈ p and 𝑥 ∈ hR,𝑟𝑒𝑔. Note that 𝐾 ′0(𝑥) = Id for all 𝑥 ∈ hR,𝑟𝑒𝑔.
Let 𝑀 > 0. By Lemma 4.3.4.1, there exists an integer 𝑁 ≥ 0, which may be taken
to be 0 if 𝑀 is sufficiently small, such that for all 𝑐 with |𝑐𝑠| < 𝑀 for all 𝑠 ∈ 𝑆 the
function 𝛿2𝑁𝐾 ′𝑐 is locally integrable on hR. As 𝛿2𝑁𝐾 ′
𝑐 is homogeneous by the same
lemma, it follows that integration against 𝛿2𝑁𝐾 ′𝑐 determines a tempered distribution
on hR. In particular, we may define a sesquilinear pairing 𝛾𝑐 on 𝛿𝑁∆𝑐(𝜆) × 𝛿𝑁∆𝑐†(𝜆)
as follows:
𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄) :=
∫hR
𝑄(𝑥)†𝛿2𝑁(𝑥)𝐾 ′𝑐(𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥 for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆.
Let 𝑈𝑀 ⊂ p be the open subset 𝑈𝑀 := 𝑐 ∈ p : |𝑐𝑠| < 𝑀 for all 𝑠 ∈ 𝑆. For any
𝑃,𝑄 ∈ C[h] ⊗ 𝜆, we see that 𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄) is a holomorphic function of 𝑐 ∈ 𝑈𝑀 ,
and the same is true for 𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄). In particular, for any such 𝑃 and 𝑄 such
that 𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄) is not identically zero for 𝑐 ∈ 𝑈𝑀 , we may define the meromorphic
function 𝑓𝑃,𝑄 : 𝑈𝑀 → C by
𝑓𝑃,𝑄(𝑐) := 𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄)/𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄).
166
By Theorem 4.3.4.7 and the properties of 𝐾 ′𝑐 listed in the previous paragraph, we see
that, for those 𝑐 ∈ p with |𝑐𝑠| sufficiently small for all 𝑠 ∈ 𝑆, the pairings 𝛾𝑐 and 𝛾𝑐agree up to a nonzero complex scalar multiple. It follows that all of the functions
𝑓𝑃,𝑄 for 𝑃,𝑄 ∈ C[h]⊗𝜆 coincide in a neighborhood of 0 in p and hence coincide with
a single meromorphic function 𝑓 : 𝑈𝑀 → C. Let 𝐾𝑐 : hR,𝑟𝑒𝑔 → EndC(𝜆) be defined
by 𝐾𝑐(𝑥) = 𝑓(𝑐)𝐾 ′𝑐(𝑥). Then 𝐾𝑐(𝑥) is meromorphic in 𝑐 ∈ 𝑈𝑀 , holomorphic in 𝑐 in a
neighborhood of 0 in p, satisfies the remaining properties of 𝐾 ′𝑐(𝑥) from the previous
paragraph, and
𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄) =
∫hR
𝑄(𝑥)†𝛿2𝑁(𝑥)𝐾𝑐(𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥 for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆
for all 𝑐 ∈ 𝑈𝑀 for which 𝐾𝑐(𝑥) is holomorphic.
It remains to show that 𝐾𝑐(𝑥) is in fact holomorphic for all 𝑐 ∈ 𝑈𝑀 . To this end,
it suffices to show that 𝐾𝑐(𝑥) has no singularities for 𝑐 in 𝑈𝑀 ∩ l for any complex line
l in p. By the Weierstrass theorem on the existence of meromorphic functions with
prescribed zeros and poles, there exists a meromorphic function 𝑔 on 𝑈𝑀 ∩ l such
that 𝐾𝑐(𝑥0) = 𝑔(𝑐)𝐾 ′′𝑐 (𝑥0), where 𝐾 ′′
𝑐 (𝑥0) ∈ EndC(𝜆) is nonzero for all 𝑐 ∈ 𝑈𝑀 ∩ l.
As above, we may extend 𝐾 ′′𝑐 (𝑥0), via 𝑊 -equivariance and the system of differential
equations (4.3.2), to a non-vanishing function 𝐾 ′′ : (𝑈𝑀 ∩ l) × hR,𝑟𝑒𝑔 → EndC(𝜆),
holomorphic in 𝑐 ∈ 𝑈𝑀 × l and analytic in 𝑥 ∈ hR,𝑟𝑒𝑔, so that 𝐾𝑐(𝑥) = 𝑔(𝑐)𝐾 ′′𝑐 (𝑥) for
all 𝑐 ∈ 𝑈𝑀 ∩ l and 𝑥 ∈ hR,𝑟𝑒𝑔. As above, we have
𝛾𝑐(𝛿𝑁𝑃, 𝛿𝑁𝑄) = 𝑔(𝑐)
∫hR
𝑄(𝑥)†𝛿2𝑁(𝑥)𝐾 ′′𝑐 (𝑥)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥 for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆
(4.3.12)
for all 𝑐 ∈ 𝑈𝑀 ∩ l for which 𝑔(𝑐) is holomorphic. But by the density of C[h]𝑒−|𝑥|2/2
in the Schwartz space S (hR) and the non-vanishing of 𝛿2𝑁(𝑥)𝐾 ′′𝑐 (𝑥) for 𝑥 ∈ hR,𝑟𝑒𝑔, it
follows that for any 𝑐 ∈ 𝑈𝑀 ∩ l there are 𝑃,𝑄 ∈ C[h] ⊗ 𝜆 such that the integral on
the righthand side of equation (4.3.12) is nonzero. It follows that 𝑔(𝑐) is holomorphic
for all 𝑐 ∈ 𝑈𝑀 ∩ l, as needed.
Corollary 4.3.5.2. a) Let 𝑞 : 𝑆 → C× be a 𝑊 -invariant function satisfying |𝑞𝑠| = 1
167
for all 𝑠 ∈ 𝑆. Then every irreducible representation of the Hecke algebra H𝑞(𝑊 )
admits a nondegenerate 𝐵𝑊 -invariant Hermitian form, unique up to R×-scaling.
b) Let 𝑐 ∈ pR and let 𝑥0 ∈ hR,𝑟𝑒𝑔. If 𝜆 ∈ Irr(𝑊 ) is such that 𝐾𝑍𝑥0(𝐿𝑐(𝜆)) is
nonzero and unitary, i.e. it admits a positive-definite 𝐵𝑊 -invariant Hermitian form,
then 𝐿𝑐(𝜆) is quasi-unitary in the sense of Definition 4.2.5.3.
Remark 4.3.5.3. Chlouveraki-Gordon-Griffeth [12, Section 4] have considered cer-
tain invariant symmetric forms on representations of the Hecke algebra H𝑞(𝑊 ) for
arbitrary 𝑞.
Proof of Corollary 4.3.5.2. Any 𝑞 as in (a) is of the form 𝑞𝑠 = 𝑒−2𝜋𝑖𝑐𝑠 for some
𝑐 ∈ pR. Fix such 𝑐 ∈ pR and choose a point 𝑥0 ∈ hR,𝑟𝑒𝑔. Any irreducible repre-
sentation of H𝑞(𝑊 ) is isomorphic to some 𝐾𝑍𝑥0(𝐿𝑐(𝜆)) for some 𝜆 ∈ Irr(𝑊 ) such
that Supp(𝐿𝑐(𝜆)) = h. Let 𝑁𝑐(𝜆) be the maximal proper submodule of ∆𝑐(𝜆), so
that 𝐿𝑐(𝜆) = ∆𝑐(𝜆)/𝑁𝑐(𝜆). Recall from Proposition 4.3.1.4(iv) that 𝑁𝑐(𝜆) is the
radical of 𝛾𝑐.
Let 𝐾𝑐 : hR,𝑟𝑒𝑔 → EndC(𝜆) be as in Theorem 4.3.5.1, so that in particular 𝐾𝑐(𝑥0)
defines a 𝐵𝑊 -invariant Hermitian form on 𝜆 =𝑣.𝑠. 𝐾𝑍𝑥0(∆𝑐(𝜆)). It follows from the
system of differential equations (4.3.2) that if 𝐾𝑐(𝑥0) = 0 then 𝐾𝑐(𝑥) = 0 for all
𝑥 ∈ hR,𝑟𝑒𝑔. In this case, by Theorem 4.3.5.1, for sufficiently large 𝑁 > 0, we have
that the restriction of 𝛾𝑐 to 𝛿𝑁∆𝑐(𝜆) is zero. As 𝛿 ∈ C[h] is self-adjoint with respect
to the form 𝛾𝑐, it follows that 𝛿2𝑁∆𝑐(𝜆) ⊂ 𝑁𝑐(𝜆), so 𝛿2𝑁𝐿𝑐(𝜆) = 0, contradicting
Supp(𝐿𝑐(𝜆)) = h. In particular, 𝐾𝑐(𝑥0) defines a nonzero 𝐵𝑊 -invariant Hermitian
form on 𝐾𝑍𝑥0(∆𝑐(𝜆)).
As 𝐾𝑍𝑥0 is exact, we have 𝐾𝑍𝑥0(𝐿𝑐(𝜆)) ∼= 𝐾𝑍𝑥0(∆𝑐(𝜆))/𝐾𝑍𝑥0(𝑁𝑐(𝜆)). Fur-
thermore, as 𝐾𝑍𝑥0(𝐿𝑐(𝜆)) is irreducible, to show that it admits a nondegenerate
𝐵𝑊 -invariant Hermitian form, unique up to R× scaling, it suffices to show that
𝐾𝑍𝑥0(𝑁𝑐(𝜆)) lies in the radical of𝐾𝑐(𝑥0). To this end, take a vector 𝑣 ∈ 𝐾𝑍𝑥0(𝑁𝑐(𝜆)).
Viewing 𝐾𝑍𝑥0(𝑁𝑐(𝜆)) as a submodule of 𝐾𝑍𝑥0(∆𝑐(𝜆)) and identifying the latter with
𝜆 as a vector space, let 𝑣 ∈ 𝑁𝑐(𝜆) ⊂ ∆𝑐(𝜆) = C[h] ⊗ 𝜆 be such that its value at 𝑥0 is
168
𝑣. In particular, 𝛿𝑁𝑣 is in the radical of 𝛾𝑐, and it follows from Theorem 4.3.5.1 that
0 = 𝛾𝑐(𝛿𝑁𝑣, 𝛿𝑁𝑄) =
∫hR
𝑄(𝑥)†𝛿2𝑁(𝑥)𝐾𝑐(𝑥)𝑣𝑒−|𝑥|2/2𝑑𝑥 for all 𝑄 ∈ C[h] ⊗ 𝜆.
As 𝛿2𝑁𝐾𝑐𝑣 defines a tempered distribution on hR with values in 𝜆, the density of
C[h]𝑒−|𝑥|2/2 in the Schwartz space S (hR) implies that 𝛿2𝑁𝐾𝑐𝑣 = 0 pointwise on
hR,𝑟𝑒𝑔. In particular, 𝐾𝑐(𝑥0)𝑣 = 0. As 𝑣 ∈ 𝐾𝑍𝑥0(𝑁𝑐(𝜆)) was arbitrary, it fol-
lows that 𝐾𝑍𝑥0(𝑁𝑐(𝜆)) ⊂ rad(𝐾𝑐(𝑥0)), where rad(𝐾𝑐(𝑥0)) denotes the radical of
the form 𝐾𝑐(𝑥0). In particular, 𝐾𝑐(𝑥0) descends to a nondegenerate 𝐵𝑊 -invariant
Hermitian form on 𝐾𝑍𝑥0(∆𝑐(𝜆)), proving (a). Note that this also shows that in fact
𝐾𝑍𝑥0(𝑁𝑐(𝜆)) = rad(𝐾𝑐(𝑥0)).
For (b), if 𝐾𝑍𝑥0(𝐿𝑐(𝜆)) is unitary, it follows from the argument above that 𝐾𝑐(𝑥0),
appropriately scaled, is positive-semidefinite and descends to a positive-definite Her-
mitian form on 𝐾𝑍𝑥0(𝐿𝑐(𝜆)). The same is then true for all points 𝑥 ∈ hR,𝑟𝑒𝑔, and the
integral formula in Theorem 4.3.5.1 then implies that the restriction of 𝛾𝑐 to 𝛿𝑁𝐿𝑐(𝜆)
is positive-definite. As 𝐿𝑐(𝜆) has full support in h and as 𝐿𝑐(𝜆)/𝛿𝑁𝐿𝑐(𝜆) is supported
on the reflection hyperplanes, it follows that the Hilbert polynomial of 𝐿𝑐(𝜆) is of
strictly higher degree than the Hilbert polynomial of 𝐿𝑐(𝜆)/𝛿𝑁𝐿𝑐(𝜆). In particular,
the Hilbert polynomial of 𝛿𝑁𝐿𝑐(𝜆) has the same leading term as the Hilbert polyno-
mial of 𝐿𝑐(𝜆) itself, so we have
lim𝑛→∞
dim(𝛿𝑁𝐿𝑐(𝜆))≤𝑛
dim𝐿𝑐(𝜆)≤𝑛= 1.
As 𝛾𝑐 is positive-definite on 𝛿𝑁𝐿𝑐(𝜆), we see that 𝐿𝑐(𝜆) is quasi-unitary, as needed.
We gave the proof of Corollary 4.3.5.2(a) as above to demonstrate a use of the
Dunkl weight function and because the arguments appearing in that proof would be
repeated later for the proof of Theorem 4.4.0.1. However, there is an alternate proof
of Corollary 4.3.5.2(a) that is also valid for finite complex reflection groups, due to
Raphaël Rouquier and communicated to me by Pavel Etingof, as follows.
Proposition 4.3.5.4. Corollary 4.3.5.2(a) is valid for finite complex reflection groups
169
𝑊 .
Proof. Let 𝑐 ∈ pR be a real parameter corresponding to 𝑞 via the KZ functor. Let
𝐿 be an irreducible representation of H𝑞(𝑊 ), and let 𝜆 be an irreducible represen-
tation of 𝑊 such that 𝐿𝑐(𝜆) has full support and 𝐿 ∼= 𝐾𝑍(𝐿𝑐(𝜆)). To show that
𝐿 admits a nondegenerate 𝐵𝑊 -invariant Hermitian form, necessarily uniquely deter-
mined up to R×-multiple, it suffices to show that 𝐿 is isomorphic to its Hermitian
dual 𝐿ℎ as a 𝐵𝑊 -representation. By Remark 4.3.4.6, there is a holomorphic func-
tion 𝐵 : p → EndC(𝜆) satisfying the conditions of Lemma 4.3.4.5. Restricting to
the complex line in p spanned by 𝑐 and scaling 𝐵 by an appropriate meromorphic
function, we may assume that 𝐵(𝑐) is nonzero and therefore determines a nonzero
𝐵𝑊 -invariant Hermitian form on 𝐾𝑍(∆𝑐(𝜆)), which we will regard as a nonzero map
𝛽 : 𝐾𝑍(∆𝑐(𝜆)) → 𝐾𝑍(∆𝑐(𝜆))ℎ. As 𝛽 defines a Hermitian form, it factors through
an isomorphism 𝛽 : 𝐾𝑍(∆𝑐(𝜆))/ ker 𝛽 ∼= (𝐾𝑍(∆𝑐(𝜆))/ ker 𝛽)ℎ. The head of the mod-
ule 𝐾𝑍(∆𝑐(𝜆)) is 𝐾𝑍(𝐿𝑐(𝜆)); this follows from the fact that 𝐿𝑐(𝜆) has full support
and is the head of ∆𝑐(𝜆) and the fact that the KZ functor admits a right adjoint
𝜋* : H𝑞(𝑊 )-mod𝑓.𝑑. → 𝒪𝑐(𝑊, h) such that 𝐾𝑍 ∘ 𝜋* ∼= IdH𝑞(𝑊 )-mod𝑓.𝑑.. In particular,
𝐾𝑍(𝐿𝑐(𝜆)) is also the head of 𝐾𝑍(∆𝑐(𝜆))/ ker 𝛽, and it appears with multiplicity 1 as
a composition factor because the same is true for 𝐿𝑐(𝜆) in ∆𝑐(𝜆). If 𝐾𝑍(∆𝑐(𝜆))/ ker 𝛽
is irreducible, then it isomorphic to 𝐾𝑍(𝐿𝑐(𝜆)), and the proof is complete. Other-
wise, let 𝑆 ⊂ 𝐾𝑍(∆𝑐(𝜆))/ ker 𝛽 be a simple submodule, necessarily not isomorphic
to 𝐾𝑍(𝐿𝑐(𝜆)) because the latter appears with multiplicity 1. Taking the Hermitian
dual, we obtain a surjection (𝐾𝑍(∆𝑐(𝜆))/ ker 𝛽)ℎ → 𝑆ℎ, and composing with 𝛽 we
conclude that 𝑆ℎ appears in the head of 𝐾𝑍(∆𝑐(𝜆))/ ker 𝛽, so 𝑆ℎ ∼= 𝐾𝑍(𝐿𝑐(𝜆)).
But 𝑆 must be isomorphic to 𝐾𝑍(𝐿𝑐(𝜇)) for some 𝜇 ∈ Irr(𝑊 ), 𝜇 = 𝜆, such that
𝐿𝑐(𝜇) has full support and appears as a composition factor in ∆𝑐(𝜆). By induction
on the highest weight order ≤𝑐 of 𝒪𝑐(𝑊, h), we may assume that 𝑆 ∼= 𝐾𝑍(𝐿𝑐(𝜇))
admits a nondegenerate 𝐵𝑊 -invariant Hermitian form, so that 𝑆 ∼= 𝑆ℎ. But then
𝑆 ∼= 𝐾𝑍(𝐿𝑐(𝜆)), a contradiction. It follows that 𝐾𝑍(∆𝑐(𝜆))/ ker 𝛽 is irreducible and
isomorphic to 𝐾𝑍(𝐿𝑐(𝜆)), and the claim follows.
170
Remark 4.3.5.5. In Section 4.4, using techniques from semiclassical analysis, we
will substantially generalize part (b) of Corollary 4.3.5.2. In particular, we will see
that, whenever 𝐿𝑐(𝜆) has full support, the asymptotic signature 𝑎𝑐,𝜆 of 𝐿𝑐(𝜆) equals,
up to sign, the signature of an invariant Hermitian form on 𝐾𝑍𝑥0(𝐿𝑐(𝜆)) normalized
by its dimension. In particular, unitarity of 𝐾𝑍𝑥0(𝐿𝑐(𝜆)) does not only imply quasi-
unitarity of 𝐿𝑐(𝜆), as in part (b) of Corollary 4.3.5.2, but is equivalent to it.
4.3.6 Extension to Tempered Distribution via the Wonderful
Model
In this section we will complete the proof of the existence of the Dunkl weight function,
i.e. of Theorem 4.3.3.1. As we will see, essentially what remains is to extend the
function 𝐾𝑐 : hR,𝑟𝑒𝑔 → EndC(𝜆) from Theorem 4.3.5.1 to a tempered distribution on
hR in a natural way. To achieve this we will use a known approach (see, e.g., [4]) in
which the extension is carried out on the De Concini-Procesi wonderful model [14];
the desired extension is then obtained by pushing forward to hR. The advantage of
working in the wonderful model rather than in hR is that the hyperplane arrangement
is replaced by a normal crossings divisor, allowing the application of standard fact
that for any 𝜆 ∈ C the function |𝑥|𝑧𝜆, locally integrable for |𝑧| small, has a natural
distributional meromorphic continuation to 𝑧 ∈ C.
For 𝑅 > 0, let 𝐵𝑅(0) denote the open disk 𝐵𝑅(0) := 𝑧 ∈ C : |𝑧| < 𝑅 of radius 𝑅
in the complex plane centered at 0. For any integer 𝑁 > 0 let 𝑀𝑎𝑡𝑁×𝑁(C) denote the
space of 𝑁 × 𝑁 complex matrices, and for any 𝑎 ∈ 𝑀𝑎𝑡𝑁×𝑁(C) let Spec(𝐴) denote
the set of eigenvalues of 𝑎.
Lemma 4.3.6.1. Let 𝑅 > 0, let 𝑎 : p → 𝑀𝑎𝑡𝑁×𝑁(C) be a linear function, and let
𝐴(𝑧; 𝑐) be a holomorphic function of (𝑐, 𝑧) ∈ p × 𝐵𝑅(0), taking values in the space
𝑀𝑎𝑡𝑁×𝑁(C) of 𝑁 ×𝑁 complex matrices, such that 𝐴(0; 𝑐) = 𝑎(𝑐) for all 𝑐 ∈ p. Then
for any bounded domain 𝑈 ⊂ p there is a 𝑀𝑎𝑡𝑁×𝑁(C)-valued holomorphic function
171
𝑄(𝑧; 𝑐) of (𝑐, 𝑧) ∈ 𝑈 ×𝐵𝑅(0) such that
𝐹 (𝑧; 𝑐) = 𝑄(𝑧; 𝑐)𝑧𝑎(𝑐)
is a solution of the differential equation
𝑧𝑑𝐹 (𝑧; 𝑐)
𝑑𝑧= 𝐴(𝑧; 𝑐)𝐹 (𝑧; 𝑐) (4.3.13)
for all 𝑐 ∈ p and a fundamental solution whenever 𝑎(𝑐) has no two eigenvalues differ-
ing by a nonzero integer.
Proof. The lemma follows from a straightforward adaptation of the arguments and re-
sults appearing in [72, Section 5] as follows. Expand the function 𝐴(𝑧; 𝑐) as 𝐴(𝑧; 𝑐) =∑∞𝑛=0 𝑎𝑐,𝑛𝑧
𝑛, where the 𝑎𝑐,𝑛 are entire functions of 𝑐 ∈ p and 𝑎𝑐,0 = 𝑎(𝑐). First,
consider a formal series 𝑃 (𝑧; 𝑐) =∑∞
𝑛=0 𝑝𝑐,𝑛𝑧𝑛 with values in 𝑀𝑎𝑡𝑁×𝑁(C). A di-
rect calculation shows that the formal series 𝑃 (𝑧; 𝑐)𝑧𝑎(𝑐) satisfies differential equation
(4.3.13) if and only if
𝑝𝑐,𝑛(𝑛+ 𝑎(𝑐)) − 𝑎(𝑐)𝑝𝑐,𝑛 =𝑛∑𝑘=1
𝑎𝑐,𝑘𝑝𝑐,𝑛−𝑘 for all 𝑛 ≥ 1. (4.3.14)
Let 𝜌𝑛+𝑎(𝑐) denote the operator of right multiplication by 𝑛 + 𝑎(𝑐) and let 𝜆𝑎(𝑐) de-
note the operator of left multiplication by 𝑎(𝑐). Note that (𝜌𝑛+𝑎(𝑐) − 𝜆𝑎(𝑐))−1 is a
meromorphic 𝐺𝐿(𝑀𝑎𝑡𝑁×𝑁(C))-valued function of 𝑐 ∈ p that is holomorphic after
multiplication by the polynomial det(𝜌𝑛+𝑎(𝑐) − 𝜆𝑎(𝑐)). In particular, we may define
meromorphic functions 𝑝𝑐,𝑛, 𝑛 ≥ 0, of 𝑐 ∈ p by setting 𝑝𝑐,0 = Id and
𝑝𝑐,𝑛 := (𝜌𝑛+𝑐𝑖𝑎 − 𝜆𝑐𝑖𝑎)−1
𝑛∑𝑘=1
𝑎𝑐,𝑘𝑝𝑐,𝑛−𝑘 for all 𝑛 ≥ 1. (4.3.15)
Let 𝑞𝑐,𝑛 := 𝑘(𝑐𝑖)𝑝𝑐,𝑛 for all 𝑛 ≥ 0.
It is a standard fact that the operator 𝜌𝑛+𝑎(𝑐) − 𝜆𝑎(𝑐) is singular if and only if
𝑛+𝑎(𝑐) and 𝑎(𝑐) have an eigenvalue in common, i.e. if 𝑛 = 𝜆−𝜆′ for some eigenvalues
𝜆, 𝜆′ ∈ Spec(𝑎(𝑐)). As 𝑎(𝑐) depends linearly on 𝑐 ∈ p and 𝑈 ⊂ p is bounded, it follows
172
that there exists 𝑚 > 0 such that 𝜌𝑛+𝑎(𝑐) − 𝜆𝑎(𝑐) is invertible for all 𝑛 > 𝑚 and
𝑐 ∈ 𝑈 , where 𝑈 denotes the closure of 𝑈 . Let 𝑘(𝑐) =∏𝑚
𝑛=1 det(𝜌𝑛+𝑎(𝑐)−𝜆𝑎(𝑐)), and let
𝑞𝑐,𝑛 := 𝑘(𝑐)𝑝𝑐,𝑛 for all 𝑛 ≥ 0. Then 𝑞𝑐,𝑛 is holomorphic function of 𝑐 in a neighborhood
of the closure 𝑈 for all 𝑛 ≥ 0 and the formal series 𝑄(𝑧; 𝑐) :=∑∞
𝑛=0 𝑞𝑐,𝑛𝑧𝑛 satisfies
differential equation (4.3.13).
It now suffices to show that there exists 𝑟 > 0 such that the series 𝑄(𝑧; 𝑐) =∑∞𝑛=0 𝑞𝑐,𝑛𝑧
𝑛 converges absolutely and uniformly for (𝑐, 𝑧) ∈ 𝑈 ×𝐵𝑟(0). In particular,
it then follows that 𝑄(𝑧; 𝑐) is holomorphic for (𝑐, 𝑧) ∈ 𝑈 × 𝐵𝑟(0), and therefore also
for (𝑐, 𝑧) ∈ 𝑈 × 𝐵𝑅(0) by standard results on holomorphic dependence of solutions
on parameters and initial conditions (see, e.g., [49, Theorem 1.1]). That 𝑄(𝑧; 𝑐) is
invertible, and hence that 𝑄(𝑧; 𝑐)𝑧𝑎(𝑐) is a fundamental solution to (4.3.13), for those
𝑐 ∈ 𝑈 such that 𝑎(𝑐) has no two eigenvalues differing by a nonzero integer follows
from the equality 𝑄(0; 𝑐) = 𝑘(𝑐)Id.
The desired uniform convergence of 𝑄(𝑧; 𝑐) =∑∞
𝑛=0 𝑞𝑐,𝑛𝑧𝑛 can be shown by a
modification of the proof of [72, Theorem 5.3] as follows. As 𝐴(𝑧; 𝑐) =∑∞
𝑛=0 𝑎𝑐,𝑛𝑧𝑛 is
holomorphic on p×𝐵𝑅(0), it follows by considering the Taylor expansions of the 𝑎𝑐,𝑛
and the boundedness of 𝑈 that there is a scalar-valued formal series 𝑏(𝑧) =∑∞
𝑛=1 𝑏𝑛𝑧𝑛
defining a holomorphic function on 𝐵𝑅(0) and such that ||𝑎𝑐,𝑛|| ≤ 𝑏𝑛 for all 𝑛 ≥ 1
and 𝑐 ∈ 𝑈 . By the boundedness of 𝑈 again, it follows that there exist 𝑁 ′, 𝐶,𝐷 > 0
such that
||(𝜌𝑛+𝑎(𝑐) − 𝜆𝑎(𝑐))−1|| ≤ 𝐶 for all 𝑛 > 𝑁 ′, 𝑐 ∈ 𝑈
and
||𝑞𝑐,𝑛|| ≤ 𝐷 for all 𝑛 ≤ 𝑁 ′, 𝑐 ∈ 𝑈.
Now, consider the formal scalar-valued series 𝑚(𝑧) =∑∞
𝑛=0𝑚𝑛𝑧𝑛 defined by
𝑚𝑟 = 𝐷 for all 𝑟 ≤ 𝑁 ′
and
𝑚𝑟 = 𝐶𝑟∑𝑠=1
𝑏𝑠𝑚𝑟−𝑠 for all 𝑟 > 𝑁 ′.
173
As 𝑚(𝑧) satisfies the equation
𝑚(𝑧) = 𝐶𝑏(𝑧)𝑚(𝑧) +𝐷 +𝑁∑𝑠=1
(𝐷 − 𝐶
𝑠∑𝑡=1
𝑏𝑡𝐷)𝑧𝑠
and 𝑏(𝑧) is holomorphic near 0 and 𝑏(0) = 0, it follows that there exists 𝑟 > 0 such
that 𝑚(𝑧) is holomorphic in 𝐵𝑟(0). By construction we have ||𝑞𝑐,𝑛|| ≤ 𝑚𝑛 for all
𝑛 ≥ 0 and 𝑐 ∈ 𝑈 , so for all 𝑐 ∈ 𝑈 and 𝑧 ∈ 𝐵𝑟(0) the series 𝑄(𝑧; 𝑐) =∑∞
𝑛=0 𝑞𝑐,𝑛𝑧𝑛 is
majorized by the series 𝑚(𝑧). In particular, 𝑄(𝑧; 𝑐) =∑∞
𝑛=0 𝑞𝑐,𝑛𝑧𝑛 is absolutely and
uniformly convergent for all 𝑐 ∈ 𝑈 and 𝑧 ∈ 𝐵𝑟(0), as needed.
Lemma 4.3.6.2. Let 𝑛,𝑁 > 0 be positive integers, let 𝑅 > 0 be a positive real
number, let 𝑎1, ..., 𝑎𝑛 : p →𝑀𝑎𝑡𝑁×𝑁(C) be linear functions, and let
𝐴𝑐 =𝑛∑𝑗=1
𝑎(𝑐)𝑧−1𝑗 𝑑𝑧𝑗 + Ω𝑐
be a meromorphic 1-form on 𝐵𝑅(0)𝑛 ⊂ C𝑛 with values in 𝑀𝑎𝑡𝑁×𝑁(C) such that
(1) the form Ω𝑐 is holomorphic on 𝐵𝑅(0)𝑛 with holomorphic dependence on 𝑐 ∈ p
(2) 𝑑+ 𝐴𝑐 defines a flat connection on (𝐵𝑅(0)∖0)𝑛 for all 𝑐 ∈ p.
Then for any bounded domain 𝑈 ⊂ p there exists, for all 1 ≤ 𝑗 ≤ 𝑛, a func-
tion 𝑄𝑗(𝑧𝑗, ..., 𝑧𝑛; 𝑐) with values in 𝑀𝑎𝑡𝑁×𝑁(C), holomorphic in (𝑐, 𝑧𝑗, ..., 𝑧𝑛) ∈ 𝑈 ×
𝐵𝑅(0)𝑛−𝑗+1, such that
𝐹 (𝑧; 𝑐) := 𝑄1(𝑧1, ..., 𝑧𝑛; 𝑐)𝑧𝑎1(𝑐)1 · · ·𝑄𝑛−1(𝑧𝑛−1, 𝑧𝑛; 𝑐)𝑧
𝑎𝑛−1(𝑐)𝑛−1 𝑄𝑛(𝑧𝑛; 𝑐)𝑧𝑎𝑛(𝑐)𝑛
is a solution of the differential equation
(𝑑+ 𝐴𝑐)𝐹 (𝑧; 𝑐) = 0
for all 𝑐 ∈ 𝑈 and a fundamental solution whenever no two eigenvalues of any 𝑎𝑖(𝑐)
differ by a nonzero integer.
Proof. The claim follows from iterated applications of Lemma 4.3.6.1 and standard
174
results (e.g. [49, Theorem 1.1]) on holomorphic dependence on parameters and initial
conditions of solutions of differential equations.
We can now complete the proof of the existence of the Dunkl weight function as a
tempered distribution with values in EndC(𝜆) with holomorphic dependence on 𝑐 ∈ p:
Proof of Theorem 4.3.3.1. Let 𝐾𝑐 : hR,𝑟𝑒𝑔 → EndC(𝜆) be the family of analytic func-
tions with holomorphic dependence on 𝑐 ∈ p in the statement of Theorem 4.3.5.1.
Note that for |𝑐𝑠| sufficiently small 𝐾𝑐 already defines a holomorphic family of tem-
pered distributions on hR with the desired properties. It therefore suffices to show
that this family of distributions for |𝑐𝑠| small has a holomorphic continuation to all
𝑐 ∈ p. To see this, note that by Proposition 4.3.3.4 the distribution 𝐾𝑐 for |𝑐𝑠| small
satisfies properties (i)-(iii) of that proposition, and therefore by Proposition 4.3.3.6
is homogeneous of degree −2𝜒𝜆(∑
𝑠∈𝑆 𝑐𝑠𝑠)/ dim𝜆. It follows that the holomorphic
continuation is homogeneous of degree −2𝜒𝜆(∑
𝑠∈𝑆 𝑐𝑠𝑠)/ dim𝜆, and hence tempered,
for all 𝑐 ∈ p. The characterizing properties (i)-(iii) appearing in Proposition 4.3.3.4
then must hold for all parameters 𝑐 ∈ p for the holomorphic continuation as they
hold for |𝑐𝑠| sufficiently small. It follows from Proposition 4.3.3.4 that such a holo-
morphic continuation is a family of tempered distributions as in Theorem 4.3.3.1. So,
it remains to construct such a holomorphic continuation.
Let 𝜋 : 𝑌 → h be the De Concini-Procesi wonderful model [14] for the hyperplane
arrangement ∪𝑠∈𝑆 ker(𝛼𝑠) ⊂ h. Recall that 𝑌 is a smooth C-variety with 𝑊 -action,
𝜋 is proper and 𝑊 -equivariant, 𝜋 restricts to an isomorphism 𝜋−1(h𝑟𝑒𝑔) → h𝑟𝑒𝑔, and
𝜋−1(∪𝑠∈𝑆 ker(𝛼𝑠)) is a normal crossings divisor. Furthermore, let 𝜋R : 𝑌R → hR denote
the real locus of 𝜋, which shares the same properties as 𝜋 but for the corresponding
real hyperplane arrangement in hR. Let 𝑌𝑟𝑒𝑔 = 𝜋−1(h𝑟𝑒𝑔) and let 𝑌R,𝑟𝑒𝑔 = 𝑌𝑟𝑒𝑔 ∩ 𝑌R =
𝜋−1R (hR,𝑟𝑒𝑔).
Note that it follows from the 𝑊 -equivariance of 𝜋 and the fact that 𝜋|𝑌𝑟𝑒𝑔 : 𝑌𝑟𝑒𝑔 →
h𝑟𝑒𝑔 is an isomorphism that 𝜋 preserves stabilizers in 𝑊 , i.e. that for all 𝑑 ∈ 𝑌 we
have Stab𝑊 (𝑑) = Stab𝑊 (𝜋(𝑑)). The inclusion Stab𝑊 (𝑑) ⊂ Stab𝑊 (𝜋(𝑑)) is trivial, so
consider 𝑤 ∈ Stab𝑊 (𝜋(𝑑)). Let 𝑈 ⊂ 𝑌 be an open neighborhood of 𝑑 in 𝑌 that is
175
stable under Stab𝑊 (𝑑) and such that the 𝑊/Stab𝑊 (𝑑)-orbits of 𝑈 are disjoint. Let
𝑈𝑟𝑒𝑔 = 𝑈 ∩ 𝑌𝑟𝑒𝑔. As 𝑤(𝜋(𝑑)) = 𝜋(𝑑) it follows that 𝑤𝜋(𝑈) ∩ 𝜋(𝑈) = ∅, and hence
𝑤𝜋(𝑈𝑟𝑒𝑔) ∩ 𝜋(𝑈𝑟𝑒𝑔) = ∅ as well. It then follows that 𝑤𝑈𝑟𝑒𝑔 ∩ 𝑈𝑟𝑒𝑔 = 𝑤𝜋−1𝜋(𝑈𝑟𝑒𝑔) ∩
𝜋−1𝜋(𝑈𝑟𝑒𝑔) = ∅ as well. In particular, 𝑤 ∈ Stab𝑊 (𝑑), so Stab𝑊 (𝑑) = Stab𝑊 (𝜋(𝑑)) for
all 𝑑 ∈ 𝑌 , as claimed.
Consider the modified KZ connection
∇′𝐾𝑍 := 𝑑−
∑𝑠∈𝑆
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
𝑠 = 𝑑−∑𝑠∈𝑆
𝑐𝑠𝑑(log𝛼𝑠)𝑠 (4.3.16)
on h𝑟𝑒𝑔. Fix a point 𝑥0 ∈ 𝒞, and let 𝐹𝑐(𝑧) be a EndC(𝜆)-valued fundamental solution of
∇′𝐾𝑍 with 𝐹𝑐(𝑥0) = Id. By standard results on holomorphic dependence of solutions
on parameters, it follows that 𝐹𝑐(𝑧) is holomorphic in 𝑐 ∈ p. Regard 𝐹𝑐(𝑧) as a
multivalued function on h𝑟𝑒𝑔. As the function 𝐾𝑐 : 𝒞 → EndC(𝜆) satisfies differential
equation (4.3.2) it follows that 𝐾𝑐(𝑥) = 𝐹𝑐†(𝑥)†,−1𝐾𝑐(𝑥0)𝐹𝑐(𝑥)−1 for all 𝑐 ∈ p and 𝑥 ∈
𝒞. In fact, by the proof of the equivalence of statements (b) and (c) in Theorem 4.3.4.7,
we have that the function 𝐾𝑐 : h𝑟𝑒𝑔 → EndC(𝜆), 𝐾𝑐(𝑧) := 𝐹𝑐†(𝑧)†,−1𝐾𝑐(𝑥0)𝐹𝑐(𝑧)−1, is
single-valued and satisfies 𝐾𝑐(𝑥) = 𝐾𝑐(𝑥) for all 𝑥 ∈ hR,𝑟𝑒𝑔.
The pullback 𝜋*∇′𝐾𝑍 is a meromorphic 1-form on 𝑌 with values in EndC(𝜆) and
singularities along the components of the divisor 𝑌 ∖𝑌𝑟𝑒𝑔 = 𝜋−1(∪𝑠∈𝑆 ker(𝛼𝑠)), with
holomorphic dependence on 𝑐 ∈ p. Following the discussion in the introduction to
[15], consider the residue of 𝜋*∇′𝐾𝑍 along the components of 𝑌 ∖𝑌𝑟𝑒𝑔 as follows. As
𝑌 ∖𝑌𝑟𝑒𝑔 is a normal crossings divisor, for any point 𝑑 ∈ 𝑌R there is a local complex
coordinate system 𝑧1, ..., 𝑧𝑙 such that 𝑧1(𝑑) = · · · 𝑧𝑙(𝑑) = 0 and the divisor 𝑌 ∖𝑌𝑟𝑒𝑔 is
given near 𝑑 by the equation 𝑧1 · · · 𝑧𝑚 = 0 for some 𝑚, 0 ≤ 𝑚 ≤ 𝑙. It follows that
for each reflection 𝑠 ∈ 𝑆 the pullback 𝜋*𝛼𝑠 is of the form 𝜋*𝛼𝑠 =∏𝑚
𝑗=1 𝑧𝑘𝑗,𝑠𝑗 𝑝𝑠(𝑧) for
some integers 𝑘𝑗,𝑠 ∈ Z≥0 and regular functions 𝑝𝑠(𝑧) with 𝑝𝑠(0) = 0. In these local
coordinates, we have, by equation (4.3.16)
𝜋*∇′𝐾𝑍 = 𝑑−
∑𝑠∈𝑆
𝑐𝑠𝑑(log(𝑚∏𝑗=1
𝑧𝑘𝑗,𝑠𝑗 𝑝𝑠))𝑠 = 𝑑−
𝑚∑𝑗=1
(∑𝑠∈𝑆
𝑐𝑠𝑘𝑗,𝑠𝑠
)𝑑 log 𝑧𝑗 + Ω𝑐
176
where Ω𝑐 is a 1-form with values in EndC(𝜆), holomorphic in 𝑧 and 𝑐.
Note that this connection is of the form considered in Lemma 4.3.6.2. For some
1 ≤ 𝑗 ≤ 𝑚, consider the residue 𝑎𝑗(𝑐) :=∑
𝑠∈𝑆 𝑐𝑠𝑘𝑗,𝑠𝑠 along the hyperplane 𝑧𝑗 = 0.
Recall that the map 𝜋 is 𝑊 -equivariant. Let 𝑑′ be a generic point on the hyperplane
𝑧𝑗 = 0, and consider the stabilizer 𝑊 ′ = Stab𝑊 (𝜋(𝑑′)) = Stab𝑊 (𝑑′), a parabolic
subgroup of 𝑊 . For 𝑠 ∈ 𝑆∖𝑊 ′, we have 𝑠(𝜋(𝑑′)) = 𝜋(𝑑′) and hence 𝛼𝑠(𝜋(𝑑′)) = 0. In
particular, in this case 𝜋*𝛼𝑠(𝑑′) = 0, so 𝑘𝑗,𝑠 = 0. Now consider instead 𝑠 ∈ 𝑆 ∩𝑊 ′
and 𝑤 ∈ 𝑊 ′. The element 𝑤 ∈ 𝑊 ′ = Stab𝑊 (𝑑′) stabilizes the hyperplane 𝑧1 = 0
and has finite order, from which it follows that 𝑤*𝑧1 = 𝑧1𝑝(𝑧) for some holomorphic
function 𝑝(𝑧) with 𝑝(0) = 0. The 𝑊 -equivariance of 𝜋 then implies that 𝑘𝑗,𝑠 =
𝑘𝑗,𝑤𝑠𝑤−1 . We see that the residue 𝑎𝑗(𝑐) is a p-linear combination of sums of conjugacy
classes of reflections in 𝑊 ′. It follows that the residues 𝑎𝑗(𝑐)𝑐∈p are simultaneously
diagonalizable with eigenvalues that are linear functions of 𝑐 ∈ p.
By Lemma 4.3.6.2, for every bounded open set 𝑈 ⊂ p there is a meromorphic
function 𝐺(𝑐) of 𝑐 ∈ 𝑈 such that, in the local coordinates 𝑧1, ..., 𝑧𝑙 on 𝑌 near 𝑑 as
above, we have
𝜋*𝐹𝑐(𝑧) = 𝑄1(𝑧1, ..., 𝑧𝑙; 𝑐)𝑧𝑎1(𝑐)1 · · ·𝑄𝑙−1(𝑧𝑙−1, 𝑧𝑙; 𝑐)𝑧
𝑎𝑙−1(𝑐)𝑙−1 𝑄𝑙(𝑧𝑙; 𝑐)𝑧
𝑎𝑙(𝑐)𝑙 𝐺(𝑐)
where the functions 𝑄𝑗(𝑧𝑗, ..., 𝑧𝑙; 𝑐) are holomorphic both in 𝑧 in a neighborhood of 𝑑
in 𝑌 and also in 𝑐 ∈ 𝑈 . As the residues 𝑎𝑗(𝑐)𝑐∈p are simultaneously diagonalizable
and as 𝐾𝑐(𝑧) = 𝐹𝑐†(𝑧)†,−1𝐾(𝑥0)𝐹𝑐(𝑧)−1 is single valued on h𝑟𝑒𝑔 and holomorphic in
𝑐, it follows from the form of 𝜋*𝐹𝑐(𝑧) above that the matrix entries of 𝜋* 𝐾𝑐(𝑧) are
linear combinations of functions of the form 𝑓𝑐(𝑧)|𝑧1|𝑔1(𝑐) · · · |𝑧𝑙|𝑔𝑙(𝑐) for some linear
functions 𝑔1, ..., 𝑔𝑙 : p → C and function 𝑓𝑐(𝑧) holomorphic in both 𝑐 ∈ 𝑈 and also
𝑧 in a neighborhood of 𝑑 in 𝑌 . It is a standard fact that for 𝜆 ∈ C the function
|𝑥|𝜆 is locally integrable on R when the real part of 𝜆 satisfies Re(𝜆) > −1 and
therefore defines a homogeneous distribution for such 𝜆, and this distribution |𝑥|𝜆
has a meromorphic continuation to all 𝜆 ∈ C with (simple) poles at the negative odd
integers. As 𝑓𝑐(𝑧) is holomorphic in 𝑐 ∈ 𝑈 and holomorphic, in particular smooth,
177
in 𝑧, it follows that the function 𝜋*𝐾𝑐(𝑥) = 𝜋* 𝐾𝑐(𝑥) on 𝑌R,𝑟𝑒𝑔 has an extension,
meromorphic in 𝑐 ∈ 𝑈 , to a distribution on an open neighborhood of 𝑑 in 𝑌R. As
the point 𝑑 ∈ 𝑌R and the bounded open set 𝑈 ⊂ p were arbitrary, it follows that the
function 𝜋*𝐾𝑐(𝑥) on 𝑌R,𝑟𝑒𝑔 has an extension to a distribution on 𝑌R,𝑟𝑒𝑔 meromorphic in
𝑐 ∈ p. Taking the pushforward of this distribution along 𝜋, one obtains an extension
of the function 𝐾𝑐 to a distribution on hR that is meromorphic in 𝑐 and coincides
with 𝐾𝑐 as a distribution when |𝑐𝑠| is small for all 𝑠 ∈ 𝑆. Denote this family of
distributions also by K𝑐
It only remains to show that K𝑐 is in fact holomorphic, not merely meromorphic,
in 𝑐 ∈ p. This follows immediately from the finiteness of the Hermite coefficients of
𝐾𝑐, which are given by the Gaussian pairing 𝛾𝑐,𝜆 by analyticity, as this is so for |𝑐𝑠|
small:
∫hR
𝑄(𝑥)†K𝑐(𝑧)𝑃 (𝑥)𝑒−|𝑥|2/2𝑑𝑥 = 𝛾𝑐,𝜆(𝑃,𝑄) ∈ C for all 𝑃,𝑄 ∈ C[h] ⊗ 𝜆 and 𝑐 ∈ p.
4.4 Signatures and the KZ Functor
In this section we will apply the Dunkl weight function and methods from semiclassical
analysis to prove the following comparison theorem for signatures of irreducible repre-
sentations of rational Cherednik algebras and Hecke algebras, generalizing Corollary
4.3.5.2:
Theorem 4.4.0.1. Let 𝜆 ∈ Irr(𝑊 ) be an irreducible representation of the finite
Coxeter group 𝑊 , and let 𝑐 ∈ pR be a real parameter. If 𝐾𝑍(𝐿𝑐(𝜆)) is nonzero then
the asymptotic signature 𝑎𝑐,𝜆 of 𝐿𝑐(𝜆) is given by the formula
𝑎𝑐,𝜆 =𝑝− 𝑞
dim𝐾𝑍(𝐿𝑐(𝜆))(4.4.1)
where 𝑝 − 𝑞 is the signature, up to sign, of a 𝐵𝑊 -invariant Hermitian form on
𝐾𝑍(𝐿𝑐(𝜆)).
178
In the special case that 𝑊 is a symmetric group, Theorem 4.4.0.1 was proved by
Venkateswaran [70, Theorem 1.4] by providing exact formulas for the left and right
sides of equation (4.4.1). The generalization above to arbitrary finite Coxeter groups
was expected by Venkateswaran [70] and Etingof.
We will begin with a reminder on semiclassical analysis in Section 4.4.1, prove
a key analytic lemma in Section 4.4.2, and finally prove Theorem 4.4.0.1 in Section
4.4.3.
4.4.1 Reminder on Semiclassical Analysis
We briefly review notation and results from semiclassical analysis, referring the reader
to [75] for details.
Denote by Opwℎ the Weyl quantization on R𝑛, mapping a symbol
𝑎(𝑥, 𝜉) ∈ 𝐶∞𝑐 (R2𝑛)
to an ℎ-dependent family of operators Opwℎ (𝑎) : 𝐿2(R𝑛) → 𝐿2(R𝑛) defined as follows:
Opwℎ (𝑎)𝑓(𝑥) =
∫R2𝑛
𝑒𝑖ℎ⟨𝑥−𝑦,𝜉⟩𝑎
(𝑥+ 𝑦
2, 𝜉)𝑓(𝑦) 𝑑𝑦𝑑𝜉. (4.4.2)
More generally one can quantize ℎ-dependent symbols 𝑎(𝑥, 𝜉;ℎ) which satisfy the
following derivative bounds for all multiindices 𝛼 on R2𝑛:
|𝜕𝛼(𝑥,𝜉)𝑎(𝑥, 𝜉;ℎ)| ≤ 𝐶𝛼𝑚(𝑥, 𝜉), (𝑥, 𝜉) ∈ R2𝑛, 0 < ℎ ≤ 1, (4.4.3)
where 𝐶𝛼 is an ℎ-independent constant depending only on the multiindex 𝛼 and where
𝑚(𝑥, 𝜉) is an order function as defined in [75, §4.4.1]. The order functions we will use
here are 𝑚 ≡ 1 and
𝑚1(𝑥, 𝜉) := 1 + |𝑥|2 + |𝜉|2. (4.4.4)
As in [75, Definition 4.4.2], for any order function 𝑚 we denote by 𝑆(𝑚) the class
of (possibly ℎ-dependent) symbols 𝑎(𝑥, 𝜉;ℎ) such that 𝑎(𝑥, 𝜉;ℎ) ∈ 𝐶∞(R2𝑛) for all
179
ℎ ∈ (0, 1] and that satisfy the derivative bounds (4.4.3). The Weyl quantization Opwℎ
is defined for symbols 𝑎 ∈ 𝑆(𝑚) and has the following properties:
1. for 𝑎 ∈ 𝑆(𝑚), the operator Opwℎ (𝑎) acts continuously on the space of Schwartz
functions S (R𝑛) and also on the space of tempered distributions and S ′(R𝑛)
[75, Theorem 4.16];
2. if 𝑎 ∈ 𝑆(1), then by [75, Theorem 4.23]
sup0<ℎ≤1
‖Opwℎ (𝑎)‖𝐿2→𝐿2 <∞; (4.4.5)
3. if 𝑎 ∈ 𝑆(𝑚), 𝑏 ∈ 𝑆(𝑚), then there exists 𝑎#𝑏 such that by [75, Theorem 4.18]
Opwℎ (𝑎)Opw
ℎ (𝑏) = Opwℎ (𝑎#𝑏), 𝑎#𝑏 ∈ 𝑆(𝑚𝑚), 𝑎#𝑏 = 𝑎𝑏+ 𝒪(ℎ)𝑆(𝑚𝑚).
(4.4.6)
4. if 𝑎 ∈ 𝑆(𝑚), then by [75, (4.1.13)]
Opwℎ (𝑎)* = Opw
ℎ (𝑎); (4.4.7)
5. if 𝑎 ∈ 𝑆(1) is real-valued and satisfies 𝑎 ≥ 𝑐 > 0 for some constant 𝑐 and all
sufficiently small ℎ, then there exists ℎ0 > 0 such that by [75, Theorem 4.30]
⟨Opwℎ (𝑎)𝑓, 𝑓⟩𝐿2 ≥ 𝑐
2‖𝑓‖2𝐿2 for all 𝑓 ∈ 𝐿2(R𝑛), 0 < ℎ ≤ ℎ0. (4.4.8)
Remark 4.4.1.1. In fact, as will soon be important for the proof of Theorem 4.4.0.1,
one may also consider, for any sufficiently large positive integer 𝑘, ℎ-dependent
symbols that are only 𝐶𝑘. Specifically, for any integer 𝑘 ≥ 0 and order function
𝑚 one may also consider the symbol class 𝑆(𝑘)(𝑚) of symbols 𝑎(𝑥, 𝜉;ℎ) such that
𝑎(𝑥, 𝜉;ℎ) ∈ 𝐶𝑘(R2𝑛) and for which the derivative bounds (4.4.3) hold for all muti-
indices 𝛼 with |𝛼| ≤ 𝑘. The construction of the Weyl quantization makes sense for
such symbols (in fact for distributions as well, see [75, Theorem 4.2]), and the proofs
180
of the properties recalled above are based on only a fixed finite number of terms in
the semiclassical expansion, and therefore on only a fixed finite number of deriva-
tives of the symbol. In particular, for 𝑘 sufficiently large, the properties of the Weyl
quantization listed above are valid for symbols from the class 𝑆(𝑘)(𝑚).
The quantum harmonic oscillator
Consider the quantum harmonic oscillator in R𝑛 [75, §6.1],
𝑃 := −ℎ2∆ + |𝑥|2.
It is a nonnegative self-adjoint operator on 𝐿2(R𝑛). We have 𝑃 = Opwℎ (𝑝) where
(recalling (4.4.4))
𝑝(𝑥, 𝜉) = |𝜉|2 + |𝑥|2 ∈ 𝑆(𝑚1).
We will also use the shifted operators determined by (𝑦, 𝜂) ∈ R2𝑛
𝑃(𝑦,𝜂) =𝑛∑𝑗=1
(−𝑖ℎ𝜕𝑥𝑗 − 𝜂𝑗)2 + |𝑥− 𝑦|2;
𝑃(𝑦,𝜂) = Opwℎ (𝑝(𝑦,𝜂)), 𝑝(𝑦,𝜂)(𝑥, 𝜉) = |𝜉 − 𝜂|2 + |𝑥− 𝑦|2 ∈ 𝑆(𝑚1).
(4.4.9)
Note that 𝑃(𝑦,𝜂) is conjugate to 𝑃 :
𝑃(𝑦,𝜂) = 𝑇(𝑦,𝜂)𝑃𝑇−1(𝑦,𝜂), 𝑇(𝑦,𝜂)𝑓(𝑥) = 𝑒
𝑖ℎ⟨𝑥,𝜂⟩𝑓(𝑥− 𝑦). (4.4.10)
We use the following consequence of functional calculus of pseudodifferential opera-
tors, see [75, Theorem 14.9] and [18, §8]:
Lemma 4.4.1.2. Assume that 𝜙 ∈ 𝐶∞𝑐 (R) is ℎ-independent and fix (𝑦, 𝜂) ∈ R2𝑛.
Then 𝜙(𝑃(𝑦,𝜂)) = Opwℎ (𝑎) for some 𝑎 satisfying for all 𝑁
𝑎 ∈ 𝑆(𝑚−𝑁1 ), 𝑎 = 𝜙 ∘ 𝑝(𝑦,𝜂) + 𝒪(ℎ)𝑆(𝑚−𝑁
1 ).
181
4.4.2 A Key Lemma
Before proving the main lemma, we will need the following elementary measure the-
oretic result:
Lemma 4.4.2.1. For any Lebesgue measurable set 𝑋 ⊂ R𝑛 let vol(𝑋) denote its
volume. For any compact set 𝐾 ⊂ R𝑛 and every 𝜖, 𝑟 > 0, there exist finitely many
points 𝑥1, ..., 𝑥𝑁 ∈ R𝑛 and positive real numbers 𝑟𝑖 ∈ (0, 𝑟) such that
(a) 𝐾 ⊂⋃𝑁𝑖=1𝐵𝑟𝑖(𝑥𝑖)
(b)∑𝑁
𝑖=1 vol(𝐵𝑟𝑖(𝑥𝑖)) < vol(𝐾) + 𝜖.
Proof. Fix the compact set 𝐾 ⊂ R𝑛 and positive numbers 𝜖, 𝑟 > 0. Recall that the
outer measures defined by coverings by rectangles or balls coincide, and that each
gives rise to Lebesgue measure. In particular, there are countably many open rect-
angles 𝑅𝑖 such that 𝐾 ⊂⋃∞𝑖=1𝑅𝑖 and
∑∞𝑖=1 vol(𝑅𝑖) < vol(𝐾) + 𝜖/2. Subdividing and
slightly enlarging the rectangles if necessary, we may further assume that vol(𝑅𝑖) <
vol(𝐵𝑟/2(0)). Similarly, there are countably many open balls 𝐵𝑟𝑖𝑗(𝑥𝑖𝑗)∞𝑗=1 such that
𝑅𝑖 ⊂ ∪∞𝑗=1𝐵𝑟𝑖𝑗(𝑥𝑖𝑗) and
∑∞𝑗=1 vol(𝐵𝑟𝑖𝑗(𝑥𝑖𝑗)) < vol(𝑅𝑖𝑗) + min𝜖/2𝑖+1, vol(𝐵𝑟/2(0)). In
particular, for each 𝑖, 𝑗 we have vol(𝐵𝑟𝑖𝑗(𝑥𝑖𝑗)) < 2vol(𝐵𝑟/2(0)) ≤ vol(𝐵𝑟(0)) so 𝑟𝑖𝑗 < 𝑟.
Also, the countably many open balls 𝐵𝑟𝑖𝑗(𝑥𝑖𝑗)∞𝑖,𝑗=1 cover the compact set 𝐾, and
we may extract a finite subcover 𝐵𝑟𝑖(𝑥𝑖)𝑁𝑖=1 of 𝐾. We have∑𝑁
𝑖=1 vol(𝐵𝑟𝑖(𝑥𝑖)) <∑∞𝑖,𝑗=1 vol(𝐵𝑟𝑖𝑗(𝑥𝑖𝑗)) < vol(𝐾) +
∑∞𝑖=1 𝜖/2
𝑖 = vol(𝐾) + 𝜖, as needed.
For 𝛽 ≥ 0, denote by 𝑉ℎ(𝛽) the range of the spectral projection 1[0,𝛽](𝑃 ):
𝑉ℎ(𝛽) = span𝑢 ∈ 𝐿2(R𝑛) | 𝑃𝑢 = 𝜆𝑢 for some 𝜆 ∈ [0, 𝛽] ⊂ 𝐿2(R𝑛).
By Weyl’s Law [75, Theorem 6.3] we have for any fixed 𝛽
dim𝑉ℎ(𝛽) = 𝑐𝑛𝛽𝑛ℎ−𝑛 + 𝑜(ℎ−𝑛) as ℎ→ 0, 𝑐𝑛 :=
1
2𝑛 · 𝑛!. (4.4.11)
Denote by
Πℎ : 𝐿2(R𝑛) → 𝐿2(R𝑛)
182
the orthogonal projector onto 𝑉ℎ(1). It is a direct consequence of [75, Theorem 6.2]
on the eigenfunctions of 𝑃 that we have
𝑉ℎ(1) = C[R𝑛]≤12( 1ℎ−𝑛)𝑒−|𝑥|2/2ℎ, (4.4.12)
where C[R𝑛]≤12( 1ℎ−𝑛) denotes the space of complex valued polynomials on R𝑛 with
degree at most 12( 1ℎ− 𝑛).
We will denote by Mat𝑚×𝑚(C) and Herm𝑚×𝑚(C) the spaces of 𝑚 × 𝑚 complex
matrices and 𝑚×𝑚 complex Hermitian matrices, respectively.
Lemma 4.4.2.2. Let 𝑚,𝑛 be positive integers, let 𝑘 >> 𝑛, and let 𝑞 : R𝑛 →
Herm𝑚×𝑚(C) be a 𝐶𝑘 function taking values in Hermitian 𝑚×𝑚 matrices and with
matrix entries 𝑞𝑖𝑗(𝑥) satisfying, for some fixed nonnegative integer 𝑁0, the derivative
bounds:
|𝜕𝛼𝑞𝑖𝑗(𝑥)| ≤ 𝐶𝛼(1 + |𝑥|)𝑁0 (4.4.13)
for all multiindices 𝛼 with |𝛼| ≤ 𝑘. Consider the family of operators, for ℎ > 0,
𝑄 = 𝑄(ℎ) : 𝑉ℎ(1) ⊗ C𝑚 → 𝑉ℎ(1) ⊗ C𝑚, 𝑄(𝑓) = (Πℎ ⊗ Id)(𝑞𝑓).
Let Spec(𝑞(𝑥)) denote the set of eigenvalues of 𝑞(𝑥) with multiplicity, and put (here
𝐵𝑟(𝑥, 𝜉) denotes the open ball with radius 𝑟 and center (𝑥, 𝜉) in R2𝑛)
𝛾 :=
∫𝐵1(0)
#𝜆 ∈ Spec(𝑞(𝑥)) : 𝜆 ≤ 0𝑑𝑥𝑑𝜉vol(𝐵1(0))
.
Then, for every 𝜖 > 0 there exists ℎ0 > 0 such that for 0 < ℎ < ℎ0 the number
of non-positive eigenvalues of 𝑄(ℎ), counted with multiplicity, is bounded above by
(𝛾 + 𝜖)𝑐𝑛ℎ−𝑛.
Proof. We follow the idea of the proof of [75, Theorem 6.8].
First, note that the operator 𝑄(ℎ) is well-defined; it follows from (4.4.12) and the
polynomial growth (4.4.13) of the matrix entries 𝑞𝑖𝑗 that multiplication by 𝑞 defines
183
a map
𝑞 : 𝑉ℎ(1) ⊗ C𝑚 → 𝐿2(R𝑛) ⊗ C𝑚,
so 𝑄(ℎ)(𝑓) = (Πℎ ⊗ Id)(𝑞𝑓) does define a linear operator on the finite-dimensional
space 𝑉ℎ(1).
Given 𝑞 and 𝜖 > 0, we may choose finitely many open balls
𝐵𝑗 := 𝐵𝑟𝑗((𝑦𝑗, 𝜂𝑗)) ⊂ R2𝑛, (𝑦𝑗, 𝜂𝑗) ∈ R2𝑛, 𝑟𝑗 > 0; 𝑗 = 1, . . . , 𝑁,
vectors 𝑣1, ..., 𝑣𝑁 ∈ C𝑚, and constants 𝑐0, 𝐶1, ..., 𝐶𝑁 > 0 all such that
𝑞(𝑥) +𝑁∑𝑗=1
𝐶𝑗1𝐵𝑟𝑗 ((𝑦𝑗 ,𝜂𝑗))(𝑥, 𝜉)𝑣𝑗𝑣
𝑇𝑗 ≥ 3𝑐0 for all (𝑥, 𝜉) ∈ 𝐵1(0) (4.4.14)
∑𝑗
vol(𝐵𝑗) ≤(𝛾 +
𝜀
3
)vol(𝐵(0, 1)), (4.4.15)
where the inequality in the first condition indicates that all eigenvalues are at least
3𝑐0 and where 1𝐵𝑟𝑗 ((𝑦𝑗 ,𝜂𝑗))denotes the indicator function of 𝐵𝑟𝑗((𝑦𝑗, 𝜂𝑗)). To see
that this is possible, we use Lemma 4.4.2.1 and descending induction on 𝑀 :=
max(𝑥,𝜉)∈𝐵1(0) #𝜆 ∈ Spec(𝑞(𝑥)) : 𝜆 ≤ 0. If 𝑀 = 0, there is nothing to prove.
Otherwise, consider the compact set
𝐾𝑀 := (𝑥, 𝜉) ∈ 𝐵1(0) : #𝜆 ∈ Spec(𝑞(𝑥)) : 𝜆 ≤ 0 = 𝑀.
For each point (𝑥, 𝜉) ∈ 𝐾𝑀 , let 𝑣(𝑥,𝜉) ∈ C𝑚 be an eigenvector of 𝑞(𝑥) with eigen-
value 𝜆(𝑥,𝜉) ≤ 0, and let 𝐶(𝑥,𝜉) = 1 − 𝜆(𝑥,𝜉). Then for all (𝑥, 𝜉) ∈ 𝐾𝑀 the ma-
trix 𝑞(𝑥) + 𝐶(𝑥,𝜉)𝑣(𝑥,𝜉)𝑣𝑇(𝑥,𝜉) has at most 𝑀 − 1 nonpositive eigenvalues counted with
multiplicity, and it follows that there exists a number 𝑟(𝑥,𝜉) > 0 such that for all
(𝑦, 𝜂) ∈ 𝐵𝑟(𝑥,𝜉)((𝑥, 𝜉)) the matrix 𝑞(𝑦) + 𝐶(𝑥,𝜉)𝑣(𝑥,𝜉)𝑣𝑇(𝑥,𝜉) has at most 𝑀 − 1 nonpos-
itive eigenvalues counted with multiplicity. As 𝐾𝑀 is compact, the open covering
𝐵𝑟(𝑥,𝜉)((𝑥, 𝜉))(𝑥,𝜉)∈𝐾𝑀of 𝐾𝑀 admits a Lebesgue number 𝑟 > 0. Applying Lemma
4.4.2.1 to 𝐾𝑀 for this 𝑟 and replacing 𝑞1 by a function defined similarly to the ex-
184
pression appearing in (4.4.14), we may reduce to the case that max(𝑥,𝜉)∈𝐵1(0) #𝜆 ∈
Spec(𝑞1(𝑥, 𝜉)) : 𝜆 ≤ 0 < 𝑀 and continue in this manner by induction, defining 𝐾𝑀−1
to be the closure of (𝑥, 𝜉) ∈ 𝐵1(0) : #𝜆 ∈ Spec(𝑞(𝑥)) : 𝜆 ≤ 0 = 𝑀 − 1, etc. The
ordered eigenvalues of the resulting function in (4.4.14) are then positive and clearly
lower-semicontinuous so are bounded away from 0 on the compact set 𝐵1(0), so the
eigenvalues are bounded below by 3𝑐0 for some 𝑐0 > 0.
As above, for (𝑦, 𝜂) ∈ R2𝑛 let 𝑝(𝑦,𝜂) be defined by 𝑝(𝑦,𝜂)(𝑥, 𝜉) = |𝑥− 𝑦|2 + |𝜉 − 𝜂|2,
and let 𝑝𝑗 = 𝑝(𝑦𝑗 ,𝜂𝑗). It is clear that there exist real numbers 𝑟𝑖 > 𝑟𝑖 and functions
𝜒, 𝜒1, ..., 𝜒𝑁 ∈ 𝐶∞𝑐 (R2𝑛) such that
𝑞 := (𝜒 ∘ 𝑝)2𝑞 + (1 − 𝜒 ∘ 𝑝)Id +𝑁∑𝑗=1
(𝜒𝑗 ∘ 𝑝𝑗)𝑣𝑗𝑣𝑇𝑗 ≥ 2𝑐0, (4.4.16)
[0, 1] ∩ Supp(1 − 𝜒) = ∅, Supp(𝜒𝑗) ⊂ (−∞, 𝑟2𝑗 ], (4.4.17)∑𝑗
𝑟2𝑛𝑗 ≤ 𝛾 +2𝜀
3, (4.4.18)
where Id ∈ Mat𝑚×𝑚(C) denotes the identity matrix.
By Remark 4.4.1.1, we may take 𝑘 sufficiently large so that the properties of
the Weyl quantization recalled in Section 4.4.1 hold for the symbol classes 𝑆(𝑘)(𝑚).
Furthermore, let Opwℎ act element-wise on matrices of such symbols. Define the
operator
𝑄1 := 𝜒(𝑃 )Opwℎ (𝑞)𝜒(𝑃 ) + (𝐼 − 𝜒(𝑃 ))Id +
𝑁∑𝑗=1
𝜒𝑗(𝑃𝑗)𝑣𝑗𝑣𝑇𝑗 .
Here Opwℎ (𝑞) is the multiplication operator by 𝑞; by (4.4.13) we have
𝑞 ∈ 𝑆(𝑘)(𝑚𝑀1 ) ⊗ Herm𝑚×𝑚(C)
for some𝑀 . Using Lemma 4.4.1.2 in each matrix entry, by the product formula (4.4.6)
we see that
𝑄1 = Opwℎ (𝑎) for some 𝑎 ∈ 𝑆(𝑘)(1) ⊗ Herm𝑚×𝑚(C),
𝑎 = 𝑞 + 𝒪(ℎ)𝑆(𝑘)(1)⊗Herm𝑚×𝑚(C).
185
That we may take 𝑎 ∈ 𝑆(𝑘)(1)⊗Herm𝑚×𝑚(C) rather than simply 𝑆(𝑘)(1)⊗Mat𝑚×𝑚(C)
follows from [75, (4.1.12)]. In particular, we see that 𝑄1 is a bounded self-adjoint
operator on 𝐿2(R𝑛) ⊗ C𝑚 by [75, Theorem 4.23].
As 𝑞 ≥ 2𝑐0 it follows that there exists ℎ0 > 0 such that 𝑎 ≥ 32𝑐0 for 0 < ℎ < ℎ0.
Recall that the square root of a positive definite matrix depends smoothly (in fact,
analytically) on the matrix entries. In particular, the square root 𝑏 :=√𝑎− 𝑐0 is
defined for 0 < ℎ < ℎ0 and is an element of the symbol class 𝑆(𝑘)(1) ⊗ Herm𝑚×𝑚(C).
As the operator Opwℎ (𝑏) is self-adjoint, it follows from product formula (4.4.6) that
we have
Opwℎ (𝑎− 𝑐0) = Opw
ℎ (𝑏)*Opwℎ (𝑏) + 𝒪(ℎ)𝐿2⊗C𝑚 .
As Opwℎ (𝑏)*Opw
ℎ (𝑏) is manifestly a nonnegative operator, shrinking ℎ0 if necessary
there is a constant 𝐶 > 0 such that Opwℎ (𝑎−𝑐0) ≥ −𝐶ℎ for 0 < ℎ < ℎ0. In particular,
Opwℎ (𝑎) ≥ 𝑐0 − 𝐶ℎ for 0 < ℎ < ℎ0. Taking ℎ0 smaller again if necessary, we have
Opwℎ (𝑎) ≥ 𝑐0
2for 0 < ℎ < ℎ0, (4.4.19)
i.e.
⟨𝑄1𝑓, 𝑓⟩ = ⟨Opwℎ (𝑎)𝑓, 𝑓⟩ ≥ 𝑐0
2||𝑓 ||2𝐿2⊗C𝑚 for all 𝑓 ∈ 𝐿2(R𝑛) ⊗ C𝑚, 0 < ℎ < ℎ0.
(4.4.20)
In particular this applies to all 𝑓 ∈ 𝑉ℎ(1) ⊗ C𝑚. However, for such 𝑓 we have
𝑓 = 𝜒(𝑃 )𝑓 , therefore
⟨𝑄𝑓, 𝑓⟩ + ⟨𝑄2𝑓, 𝑓⟩ = ⟨𝑄1𝑓, 𝑓⟩ ≥𝑐02‖𝑓‖2 for all 𝑓 ∈ 𝑉ℎ(1) ⊗ C𝑚 (4.4.21)
where
𝑄2 :=𝑁∑𝑗=1
𝜒𝑗(𝑃𝑗)𝑣𝑗𝑣𝑇𝑗 . (4.4.22)
and where the inner products and norms above are in 𝐿2(R𝑛) ⊗ C𝑚. By (4.4.10),
186
(4.4.11), and (4.4.18) we estimate the rank of 𝑄2 for small ℎ by
rank𝑄2 ≤𝑁∑𝑗=1
dim𝑉ℎ(𝑟2𝑗 ) = 𝑐𝑛ℎ
−𝑛𝑁∑𝑗=1
𝑟2𝑛𝑗 + 𝑜(ℎ−𝑛) ≤ (𝛾 + 𝜀)𝑐𝑛ℎ−𝑛. (4.4.23)
Together (4.4.21) and (4.4.23) give the required estimate.
4.4.3 Proof of the Signature Comparison Theorem
Proof of Theorem 4.4.0.1. Let 𝑘 be a positive integer sufficiently large so that Lemma
4.4.2.2 holds for 𝐶𝑘 Herm(𝜆)-valued functions on hR satisfying the required derivative
bounds appearing in that lemma. Let 𝑐 ∈ pR be a real parameter, and suppose
Supp𝐿𝑐(𝜆) = h. Let
𝐾𝑐 : hR,𝑟𝑒𝑔 → Herm(𝜆)
be the function appearing in Theorem 4.3.5.1, i.e. 𝐾𝑐 is the restriction of the Dunkl
weight function at parameter 𝑐 to hR,𝑟𝑒𝑔. By Theorem 4.3.5.1 there is a positive integer
𝑁 > 0 such that
𝛾𝑐(𝛿𝑁𝑃1, 𝛿
𝑁𝑃2) =
∫hR
𝑃2(𝑥)†𝛿2𝑁(𝑥)𝐾𝑐(𝑥)𝑃1(𝑥)𝑒−|𝑥|2/2𝑑𝑥 for all 𝑃1, 𝑃2 ∈ C[h] ⊗ 𝜆.
By Lemma 4.3.4.1 𝐾𝑐 is a homogeneous function and we may take 𝑁 large enough so
that 𝛿2𝑁𝐾𝑐 extends continuously to all of hR with value 0 on the union hR∖hR,𝑟𝑒𝑔 of
the reflection hyperplanes. By the homogeneity of 𝐾𝑐, it follows that we may take 𝑁
large enough so that 𝛿2𝑁𝐾𝑐 : hR → Herm(𝜆) is in fact 𝐶𝑘 and satisfies the hypotheses
of the function 𝑞 appearing in Lemma 4.4.2.2.
Let 𝑑 be the degree of homogeneity of the function 𝛿2𝑁𝐾𝑐, let ℎ > 0 be arbitrary,
and let 𝑄(ℎ) : 𝑉ℎ(1) ⊗ 𝜆→ 𝑉ℎ(1) ⊗ 𝜆, 𝑉ℎ(1) ⊂ 𝐿2(hR), be the operator
𝑄(ℎ)(𝑓) := (Πℎ ⊗ Id)((𝛿2𝑁𝐾𝑐)𝑓)
187
considered in Lemma 4.4.2.2. Recall from (4.4.12) that
𝑉ℎ(1) = C[h]≤12( 1ℎ−𝑙)𝑒−|𝑥|2/2ℎ
where 𝑙 = dim h. For any 𝑃1, 𝑃2 ∈ C[h]≤12( 1ℎ−𝑙) ⊗ 𝜆 we have
𝛾𝑐,𝜆(𝛿𝑁𝑃1, 𝛿
𝑁𝑃2) =
∫hR
𝑃2(𝑥)†(𝛿2𝑁𝐾𝑐)(𝑥)𝑃1(𝑥)𝑒−|𝑥|2/2𝑑𝑥
=
∫hR
(𝑃2(𝑥)𝑒−|𝑥|2/4)†(𝛿2𝑁𝐾𝑐)(𝑥)(𝑃1(𝑥)𝑒−|𝑥|2/4)𝑑𝑥
= (2/ℎ)𝑙/2∫hR
(𝑃2(√
2/ℎ𝑥)𝑒−|𝑥|2/2ℎ)†(𝛿2𝑁𝐾𝑐)(√
2/ℎ𝑥)(𝑃1(√
2/ℎ𝑥)𝑒−|𝑥|2/2ℎ)𝑑𝑥
= (2/ℎ)(𝑙+𝑑)/2∫hR
(𝑃2(√
2/ℎ𝑥)𝑒−|𝑥|2/2ℎ)†(𝛿2𝑁𝐾𝑐)(𝑥)(𝑃1(√
2/ℎ𝑥)𝑒−|𝑥|2/2ℎ)𝑑𝑥
= (2/ℎ)(𝑙+𝑑)/2⟨𝑄(ℎ)
(𝑃1(√
2/ℎ𝑥)𝑒−|𝑥|2/2ℎ), 𝑃2(
√2/ℎ𝑥)𝑒−|𝑥|2/2ℎ
⟩𝐿2(hR)⊗𝜆
where ⟨·, ·⟩𝐿2(hR)⊗𝜆 denotes the inner product on 𝐿2(hR)⊗𝜆 induced from the standard
inner product on 𝐿2(hR) and the inner product (·, ·)𝜆 on 𝜆. In particular, for all ℎ > 0,
maxdim𝑈 : 𝑈 ⊂ C[h]≤12( 1ℎ−𝑙) ⊗ 𝜆, 𝛾|𝛿𝑁𝑈 is positive definite
= #𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 > 0,(4.4.24)
maxdim𝑈 : 𝑈 ⊂ C[h]≤12( 1ℎ−𝑙) ⊗ 𝜆, 𝛾|𝛿𝑁𝑈 is negative definite
= #𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 < 0,(4.4.25)
and
dim rad(𝛾|𝛿𝑁C[h]≤
12 ( 1
ℎ−𝑙)⊗𝜆
)= dim ker𝑄(ℎ). (4.4.26)
For a point 𝑥 ∈ hR,𝑟𝑒𝑔, let 𝑝 be the dimension of a maximal positive-definite
subspace of 𝜆 with respect to the Hermitian form 𝐾𝑐(𝑥), let 𝑞 be the dimension
of a maximal negative-definite subspace of 𝜆 with respect to 𝐾𝑐(𝑥), and let 𝑟 =
dim rad(𝐾𝑐(𝑥)) = dim𝜆 − 𝑝 − 𝑞 be the dimension of the radical of 𝐾𝑐(𝑥). Recall
that the Hermitian forms 𝐾𝑐(𝑥) and 𝐾𝑐(𝑥′) are equivalent for all 𝑥, 𝑥′ ∈ hR,𝑟𝑒𝑔, so the
188
integers 𝑝, 𝑞, 𝑟 do not depend on the choice of the point 𝑥 ∈ hR,𝑟𝑒𝑔. As 𝐾𝑐(𝑥) descends
to a 𝐵𝑊 -invariant non-degenerate Hermitian form on 𝐾𝑍𝑥(𝐿𝑐(𝜆)), the quantity 𝑝− 𝑞
is as in the statement of Theorem 4.4.0.1.
As usual, let 𝑁𝑐(𝜆) denote the maximal proper submodule of ∆𝑐(𝜆), and recall
that 𝑁𝑐(𝜆) = rad(𝛾𝑐,𝜆). Recall also from the proof of Corollary 4.3.5.2 that for any
𝑥 ∈ hR,𝑟𝑒𝑔 we have 𝐾𝑍𝑥(𝑁𝑐(𝜆)) = rad(𝐾𝑐(𝑥)), where we make the usual identification
of 𝐾𝑍𝑥(∆𝑐(𝜆)) with 𝜆 as a vector space and the identification of 𝐾𝑍𝑥(𝑁𝑐(𝜆)) as a
subspace of 𝐾𝑍𝑥(∆𝑐(𝜆)). In particular, 𝑁𝑐(𝜆) is a graded C[h]-module whose restric-
tion to h𝑟𝑒𝑔 is an algebraic vector bundle of rank 𝑟 = dim rad(𝐾𝑐(𝑥)). It follows from
standard results on Hilbert series and equation (4.4.26) that we have
limℎ→0
dim ker𝑄(ℎ)
dim𝑉ℎ(1)= 𝑟. (4.4.27)
In the case of the function 𝛿2𝑁𝐾𝑐, the constant 𝛾 appearing in Lemma 4.4.2.2 is given
by 𝛾 = 𝑞+ 𝑟. In particular, it follows from that lemma and equation (4.4.27) that we
have
lim supℎ→0
#𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 < 0dim𝑉ℎ(1)
= 𝑞. (4.4.28)
Similarly, applying Lemma 4.4.2.2 to the function −𝛿2𝑁𝐾𝑐 we have
lim supℎ→0
#𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 > 0dim𝑉ℎ(1)
= 𝑝. (4.4.29)
As
#𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 > 0dim𝑉ℎ(1)
+#𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 < 0
dim𝑉ℎ(1)+
dim ker𝑄(ℎ)
dim𝑉ℎ(1)= dim𝜆
we see that equations (4.4.28) and (4.4.29) hold with limits replacing lim sups. We
therefore have, by equations (4.4.24) and (4.4.25),
lim𝑛→∞
𝑠𝑖𝑔𝑛(𝛾𝑐,𝜆|(𝛿𝑁𝐿𝑐(𝜆))≤𝑛)
dim(𝛿𝑁𝐿𝑐(𝜆))≤𝑛
189
= limℎ→0
#𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 > 0 − #𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 < 0#𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 > 0 + #𝜇 ∈ Spec(𝑄(ℎ)) : 𝜇 < 0
=𝑝− 𝑞
𝑝+ 𝑞
=𝑝− 𝑞
dim𝐾𝑍(𝐿𝑐(𝜆)).
As Supp𝐿𝑐(𝜆) = h we have, as explained in the proof of Corollary 4.3.5.2,
lim𝑛→∞
dim(𝛿𝑁𝐿𝑐(𝜆))≤𝑛
dim𝐿𝑐(𝜆)≤𝑛= 1.
In particular, the asymptotic signature of 𝐿𝑐(𝜆) is given by
𝑎𝑐,𝜆 = lim𝑛→∞
𝑠𝑖𝑔𝑛(𝛾𝑐,𝜆|𝐿𝑐(𝜆)≤𝑛)
dim𝐿𝑐(𝜆)≤𝑛= lim
𝑛→∞
𝑠𝑖𝑔𝑛(𝛾𝑐,𝜆|(𝛿𝑁𝐿𝑐(𝜆))≤𝑛)
dim(𝛿𝑁𝐿𝑐(𝜆))≤𝑛=
𝑝− 𝑞
dim𝐾𝑍(𝐿𝑐(𝜆)),
as claimed.
4.5 Conjectures and Further Directions
In this section we will discuss various possible extensions of the results of this chapter.
4.5.1 Complex Reflection Groups
The Dunkl weight function 𝐾𝑐,𝜆 constructed in this chapter is, at present, a phe-
nomenon for finite real reflection groups. There are at least two relevant objects
that are absent for finite complex reflection groups. First, the weight function 𝐾𝑐,𝜆
is a distribution on the real reflection representation hR, but the complex reflection
representation of a finite complex reflection group in general does not arise as the
complexification of a real reflection representation. Second, the Gaussian inner prod-
uct 𝛾𝑐,𝜆, for which 𝐾𝑐,𝜆 gives an integral formula, is itself absent for complex reflection
groups. While the contravariant form 𝛽𝑐,𝜆 is defined for an arbitrary finite complex
reflection group 𝑊 and an irreducible representation 𝜆, the definition of 𝛾𝑐,𝜆 relies on
the element f from the canonical sl2-triple e, f ,h ∈ 𝐻𝑐(𝑊, h) - but for complex 𝑊
190
there is no such sl2-triple.
Nevertheless, from the proof of Theorem 4.3.4.7, when |𝑐𝑠| is small for all 𝑠 ∈ 𝑆
the Dunkl weight function 𝐾𝑐,𝜆 is given by integration against an analytic function
𝐾 : hR,𝑟𝑒𝑔 → EndC(𝜆) that has a single-valued extension 𝐾(𝑧) to h𝑟𝑒𝑔 of the form
𝐾(𝑧) = 𝐹𝑐†(𝑧)†,−1 𝐾(𝑥0)𝐹𝑐(𝑧)−1, (4.5.1)
where 𝑥0 ∈ hR,𝑟𝑒𝑔, 𝐾(𝑥0) ∈ EndC(𝜆) defines a 𝐵𝑊 -invariant sesquilinear pairing
𝐾𝑍𝑥0(∆𝑐(𝜆)) ×𝐾𝑍𝑥0(∆𝑐†(𝜆)) → C,
and 𝐹𝑐(𝑧) is the monodromy of the modified KZ connection
∇′𝐾𝑍 = 𝑑−
∑𝑠∈𝑆
𝑐𝑠𝑑𝛼𝑠𝛼𝑠
𝑠
from 𝑥0 to 𝑧. All of these ingredients are available for finite complex reflection groups
as well. By Remark 4.3.4.6, when 𝑊 is a finite complex reflection group, for any
𝑥0 ∈ h𝑟𝑒𝑔 there is an operator 𝐴𝑐 ∈ EndC(𝜆) defining a 𝐵𝑊 -invariant sesquilinear
pairing
𝐾𝑍𝑥0(∆𝑐(𝜆)) ×𝐾𝑍𝑥0(∆𝑐†(𝜆)) → C,
unique up to scalar multiple for generic 𝑐. The modified KZ connection is naturally
generalized as
∇′𝐾𝑍 = 𝑑−
∑𝑠∈𝑆
2𝑐𝑠𝑑𝛼𝑠
(1 − 𝜆𝑠)𝛼𝑠𝑠,
where 𝜆𝑠 is the nontrivial eigenvalue of the complex reflection 𝑠 ∈ 𝑆. Taking 𝐾(𝑥0) =
𝐴𝑐 and defining 𝐾(𝑧) for 𝑧 ∈ h𝑟𝑒𝑔 as in equation (4.5.1) defines a single-valued function𝐾 : h𝑟𝑒𝑔 → EndC(𝜆). For generic 𝑐, such a function is uniquely determined up to a
global scalar multiple. It would be interesting to understand the meaning of these
functions for finite complex reflection groups and to investigate their relationship to
the contravariant form 𝛽𝑐,𝜆.
191
4.5.2 Preservation of Jantzen Filtrations, Epsilon Factors, and
Signatures
In Section 4.2.1, we introduced Jantzen filtrations for standard modules ∆𝑐(𝜆), de-
pending on a base parameter 𝑐0 ∈ pR and a deformation direction 𝑐1 ∈ pR and
arising from the analytic (in fact, polynomial) family of forms 𝛽𝑐,𝜆. Given a point
𝑥0 ∈ hR,𝑟𝑒𝑔, the Dunkl weight function provides a natural family 𝐾𝑐,𝜆(𝑥0) of 𝐵𝑊 -
invariant Hermitian forms on 𝐾𝑍𝑥0(∆𝑐(𝜆)), holomorphic in 𝑐 ∈ p. Given a base
parameter 𝑐0 ∈ pR and deformation direction 𝑐1 ∈ p, one therefore obtains a Jantzen
filtration on 𝐾𝑍𝑥0(∆𝑐(𝜆)) by H𝑞(𝑊 )-submodules. The proof of Corollary 4.3.5.2
shows that 𝐾𝑍𝑥0(rad(𝛽𝑐,𝜆)) = rad(𝐾𝑐(𝑥0)), i.e. 𝐾𝑍𝑥0 intertwines the first two terms
of the Jantzen filtration on ∆𝑐(𝜆) with the first two terms of the Jantzen filtration
on 𝐾𝑍𝑥0(∆𝑐(𝜆)). It is natural to expect that this is true for all terms in the Jantzen
filtration:
Conjecture 4.5.2.1. For any base parameter 𝑐0 ∈ pR and deformation direction 𝑐1 ∈
pR, the functor 𝐾𝑍𝑥0 sends the Jantzen filtration on ∆𝑐(𝜆) to the Jantzen filtration
on 𝐾𝑍𝑥0(∆𝑐(𝜆)).
In fact, it may be possible to adapt the argument that 𝐾𝑍𝑥0(rad(𝛽𝑐, 𝜆)) =
rad(𝐾𝑐(𝑥0)) appearing in the proof of Corollary 4.3.5.2 directly to prove Conjec-
ture 4.5.2.1. Alternatively, Ivan Losev has suggested that Conjecture 4.5.2.1 may
be able to be proved by a purely categorical argument by considering appropriate
C[[𝑡]]-linear categories deforming 𝒪𝑐(𝑊, h) and H𝑞(𝑊 )-mod𝑓.𝑑. and interpreting the
relevant Hermitian forms as maps from standard objects to (complex conjugates of)
costandard objects.
Additionally, by Theorem 4.4.0.1, the asymptotic signature of the graded Hermi-
tian form on the first subquotient of the Jantzen filtration of ∆𝑐(𝜆) equals the normal-
ized signature of the Hermitian form on the first subquotient of the Jantzen filtration
of 𝐾𝑍𝑥0(∆𝑐(𝜆)). It is natural to expect that this too holds for all of the subquotients,
although this should be significantly more difficult to show than Conjecture 4.5.2.1.
In fact, this will follow from Lemmas 4.2.4.2 and 4.2.4.3, and their analogues at the
192
level of Hecke algebra representations, if the epsilon factors 𝜖𝑖 ∈ ±1 appearing in
Lemma 4.2.4.3 are also compatible with 𝐾𝑍𝑥0 :
Conjecture 4.5.2.2. For any base parameter 𝑐0 ∈ pR and deformation direction 𝑐1 ∈
pR, Conjecture 4.5.2.1 holds, and for each 𝑘 ≥ 0, the epsilon factors attached to each
full-support irreducible constituent of the 𝑘𝑡ℎ subquotient of the Jantzen filtration on
∆𝑐(𝜆) determine epsilon factors for the irreducible constituents of the 𝑘𝑡ℎ subquotient
of the Jantzen filtration on 𝐾𝑍𝑥0(∆𝑐(𝜆)). In particular, the asymptotic signature of
the induced form on the 𝑘𝑡ℎ subquotient of the Jantzen filtration of ∆𝑐(𝜆) equals the
normalized signature, in the sense of Theorem 4.4.0.1, of the induced form on the 𝑘𝑡ℎ
subquotient of the Jantzen filtration on 𝐾𝑍𝑥0(∆𝑐(𝜆)).
Furthermore, after making an appropriate definition of Jantzen filtrations on
𝐾𝑍𝑥0(∆𝑐(𝜆)) in the complex reflection group case, perhaps following the ideas of
Section 4.5.1, one may similarly formulate Conjecture 4.5.2.2 in the complex reflec-
tion group case. One possible approach to prove such a conjecture could be to develop
an analogue for rational Cherednik algebras of the signed Kazhdan-Lusztig polyno-
mials introduced by Yee [74]. This would give an approach for providing an entirely
algebraic proof of Theorem 4.4.0.1 that is also valid for complex reflection groups.
4.5.3 Local Description of 𝐾𝑐,𝜆 and Modules of Proper Support
Let 𝑐 ∈ pR be a real parameter. Theorem 4.4.0.1 shows that, when 𝐿𝑐(𝜆) has full
support, its asymptotic signature 𝑎𝑐,𝜆 can be described in a simple way in terms of
the local nature of the distribution 𝐾𝑐,𝜆 near a generic point of its support in hR. It
is natural to expect a similar statement to hold for arbitrary 𝐿𝑐(𝜆).
From the proof of Theorem 4.3.4.7, the distribution 𝐾𝑐,𝜆, locally near any 𝑏 ∈
hR,𝑟𝑒𝑔, is given by integration against an analytic function of the form
𝐵𝑐(𝑥)†,−1𝐾𝑐,𝜆(𝑏)𝐵𝑐(𝑥)−1,
where 𝐾𝑐,𝜆(𝑏) ∈ Herm(𝜆) and 𝐵𝑐(𝑥) is an analytic 𝐺𝐿(𝜆)-valued function of 𝑥 defined
193
near 𝑥0 and satisfying 𝐵𝑐(𝑏) = Id. Rearranging, we have
𝐵𝑐(𝑥)†𝐾𝑐,𝜆(𝑥)𝐵𝑐(𝑥) = 𝐾𝑐,𝜆(𝑏) (4.5.2)
for 𝑥 ∈ hR,𝑟𝑒𝑔 near 𝑏.
Now consider an arbitrary point 𝑏 ∈ hR be an arbitrary point. Let 𝑊 ′ = Stab𝑊 (𝑏),
let h𝑊 ′ be the unique 𝑊 ′-stable complement to h𝑊′ in h, and let hR,𝑊 ′ = hR ∩ h𝑊 ′ .
The action of 𝑊 ′ on hR,𝑊 ′ realizes 𝑊 ′ as a finite real reflection group, generated by
the reflections 𝑆 ′ := 𝑆 ∩𝑊 ′. In particular, we have the rational Cherednik algebra
𝐻𝑐′(𝑊′, h𝑊 ′) and its category 𝒪𝑐′(𝑊
′, h𝑊 ′) with parameter 𝑐′ := 𝑐|𝑆′ . In a sense,
the local structure of a module 𝑀 ∈ 𝒪𝑐(𝑊, h) near the point 𝑏 ∈ h respects the
decomposition h = h𝑊′ ⊕ h𝑊 ′ , with the structure in the h𝑊 ′ component described by
the image Res𝑏𝑀 ∈ 𝒪𝑐′(𝑊′, h𝑊 ′) of 𝑀 under the Bezrukavnikov-Etingof parabolic
restriction functor Res𝑏 [5, Section 3.5] and with the structure in the h𝑊′ compo-
nent described by an associated local system [5, Section 3.7, Proposition 3.20]. The
following conjecture describing the structure of 𝐾𝑐,𝜆 near 𝑏 is a natural analogue:
Conjecture 4.5.3.1. Let 𝑐 ∈ pR be a real parameter, let 𝑏 ∈ hR be an arbitrary point,
let 𝑊 ′ = Stab𝑊 (𝑏), and let 𝑥 = (𝑥′, 𝑥′′) denote the decomposition of a vector 𝑥 ∈ hR
with respect to the direct sum decomposition hR = hR,𝑊 ′ ⊕ h𝑊′
R . Then there exists a
𝑊 ′-equivariant EndC(𝜆)-valued analytic function 𝐵(𝑥) defined in a neighborhood of 𝑏
such that 𝐵(𝑏) = Id and
𝐵(𝑥)†𝐾𝑐,𝜆(𝑥)𝐵(𝑥) =∑
𝜇∈Irr(𝑊 ′)
𝐾𝑐,𝜇(𝑥′) ⊗ ℎ𝜇, (4.5.3)
where, for each 𝜇 ∈ Irr(𝑊 ′), ℎ𝜇 is a Hermitian form on Hom𝑊 ′(𝜇, 𝜆).
A similar statement should hold for all 𝑐 ∈ p. For arbitrary parameters 𝑐 ∈ p, the
lefthand side of equation (4.5.3) should be replaced by 𝐵𝑐†(𝑥)†𝐾𝑐,𝜆(𝑥)𝐵𝑐(𝑥), and the
ℎ𝜇 should be elements of EndC(Hom𝑊 ′(𝜇, 𝜆)), not necessarily Hermitian. Generalizing
the case 𝑏 ∈ hR,𝑟𝑒𝑔 in which ℎ𝜆 gives a 𝐵𝑊 -invariant Hermitian form on the H𝑞(𝑊 )-
representation 𝜆 ∼=C 𝐾𝑍𝑏(∆𝑐(𝜆)), it is reasonable to expect that for all 𝜇 ∈ Irr(𝑊 ′)
194
the space Hom𝑊 ′(𝜇, 𝜆) is a representation of a generalized Hecke algebra, in the sense
of [52, Definition 3.25], and that the operator ℎ𝜇 is invariant under the action of the
fundamental group 𝜋1(h𝑊′
𝑟𝑒𝑔/𝐼) appearing in [52, Definition 3.25].
We can already see that Conjecture 4.5.3.1 holds in several cases. When 𝑏 ∈ hR,𝑟𝑒𝑔,
we have 𝑊 ′ = 1, and Conjecture 4.5.3.1 holds by equation (4.5.2). When 𝑏 = 0, we
have 𝑊 ′ = 𝑊 , and Conjecture 4.5.3.1 is trivial - simply take 𝐵(𝑥) = Id. When 𝑏
is a generic point on a reflection hyperplane, i.e. 𝑊 ′ = ⟨𝑠⟩ for some 𝑠 ∈ 𝑆, that
Conjecture 4.5.3.1 holds follows from the proof of Theorem 4.3.4.7 - the function
𝑃𝑖,𝑐(𝑥) gives the required function 𝐵(𝑥) and the operators 𝐾1,1𝑖 , 𝐾−1,−1
𝑖 ∈ EndC(𝜆),
viewed as Hermitian forms on Hom𝑊 ′(triv, 𝜆) and on Hom𝑊 ′(sgn, 𝜆), respectively,
give the required forms ℎtriv and ℎsgn, where triv and sgn denote the trivial and sign
representations of 𝑊 ′, respectively.
In view of Conjecture 4.5.3.1, a natural extension of Theorem 4.4.0.1 to irreducible
representations 𝐿𝑐(𝜆) of arbitrary support is the following:
Conjecture 4.5.3.2. Let 𝑐 ∈ pR be a real parameter, let 𝜆 ∈ Irr(𝑊 ) be an irreducible
representation of 𝑊 , let 𝑏 ∈ hR be a generic point in Supp(𝐿𝑐(𝜆)) ∩ hR = Supp𝐾𝑐,𝜆.
Then, using the notation from Corollary 4.5.3.1, the asymptotic signature 𝑎𝑐,𝜆 of 𝐿𝑐,𝜆
is given by
𝑎𝑐,𝜆 =
∑𝜇∈Irr(𝑊 ′),dim𝐿𝑐(𝜇)<∞ dim𝐿𝑐(𝜇)𝑎𝑐,𝜇sign(ℎ𝜇)
dim Res𝑏𝐿𝑐(𝜆). (4.5.4)
Recall that Res𝑏𝐿𝑐(𝜆) is nonzero and finite-dimensional if and only if 𝑏 is a generic
point in Supp(𝐿𝑐(𝜆)) [5, Proposition 3.23], so the righthand side of equation (4.5.4)
is well-defined. When 𝑏 = 0, we have that 𝑊 ′ = 𝑊 , ℎ𝜇 = 0 unless 𝜇 = 𝜆, ℎ𝜆 is the
standard form on Hom𝑊 ′(𝜆, 𝜆) = C, and Res𝑏𝐿𝑐(𝜆) = 𝐿𝑐(𝜆), so Conjecture 4.5.3.2
holds trivially. When 𝑏 ∈ hR,𝑟𝑒𝑔, Conjecture 4.5.3.2 is precisely Theorem 4.4.0.1, as
in that case we have 𝑊 ′ = 1, 𝐿𝑐(𝜇) = 𝐿𝑐(triv) = C (there are no other irreducible
representations in the sum), 𝑎𝑐,𝜇 = 1, and Res𝑏𝐿𝑐(𝜆) = 𝐾𝑍(𝐿𝑐(𝜆)) as vector spaces
[5, Remark 3.16].
195
196
Bibliography
[1] J. Adams, M. can Leeuwen, P. Trapa, D. Vogan. Unitary representations of realreductive groups. Preprint, arXiv:1212.2192v5.
[2] M. Balagovic, A. Puranik. Irreducible representations of the rational Cherednikalgebra associated to the Coxeter group 𝐻3. J. of Algebra 405 (2014), 259-290.
[3] Y. Berest, P. Etingof and V. Ginzburg. Finite-dimensional representations of ra-tional Cherednik algebras. Int. Math. Res. Not. 2003, no. 19, 1053-1088.
[4] C. Bergbaurer, R. Brunetti, D. Kreimer. Renormalization and resolution of sin-gularities. Preprint, arXiv:0908.0633.
[5] R. Bezrukavnikov, P. Etingof. Parabolic induction and restriction functors forrational Cherednik algebras. Selecta Math. (N.S.) 14 (2009), no. 3-4, 397-425.
[6] E. Brieskorn, K. Saito. Artin-gruppen und Coxeter-gruppen. Invent. Math. 17(1972) 245-271.
[7] M. Broué, G. Malle, R. Rouquier. Complex reflection groups, braid groups, Heckealgebras. J. reine angew. Math. 500 (1998), 127-190.
[8] R. W. Carter. Finite Groups of Lie Type: Conjugacy Classes and Complex Char-acters. John Wiley & Sons Ltd., New York (1985).
[9] E. Chavli. The Broué-Malle-Rouquier conjecture for the exceptional groups ofrank 2, Ph.D. Thesis, U. Paris Diderot, 2016.
[10] I. Cherednik. Affine Hecke algebras via DAHA, talk, available atwww-math.mit.edu/∼etingof/hadaha.pdf.
[11] C. Chevalley. Invariants of finite groups generated by reflections. Amer. J. Math77(4) (1955), 778-782.
[12] M. Chlouverkai, I. Gordon, S. Griffeth. Cell modules and canonical basic sets forHecke algebras from Cherednik algebras. Contemp. Math. Vol. 562, New Trendsin Noncommutative Algebra (2012), 77-89.
[13] T. Chmutova. Representations of the rational Cherednik algebra of dihedral type.J. of Algebra 297 (2006), 542-565.
197
[14] C. De Concini and D. Procesi. Wonderful models of subspace arrangements. Se-lecta Math. (N.S.) 1 (1995), 459-494.
[15] C. De. Concini and D. Procesi. Hyperplane arrangements and holonomy equa-tions. Selecta Math. (N.S.) 1 (1995), 495-535.
[16] P. Deligne. Equations différentielles à points singuliers réguliers. Lecture Notesin Math. 163, Springer Verlag, 1970.
[17] P. Deligne, Les immeubles de groupes de tresses generalises. Invent. Math. 17(1972), 273-302.
[18] Mouez Dimassi and Johannes Sjöstrand. Spectral asymptotics in the semi-classical limit. Cambridge University Press, 1999.
[19] C. Dunkl. Vector polynomials and a matrix weight associated to dihedral groups.SIGMA 10 (2014), 044.
[20] C. Dunkl. Vector-valued polynomials and a matrix weight function with 𝐵2-action. SIGMA 9 (2013), 007.
[21] C. Dunkl. Vector-valued polynomials and a matrix weight function with 𝐵2-action, II. SIGMA 9 (2013), 043.
[22] C. Dunkl. Orthogonality measure on the torus for vector-valued Jack polynomials.SIGMA 12 (2016), 033.
[23] C. Dunkl. A linear system of differential equations related to vector-valued Jackpolynomials on the torus. SIGMA 13 (2017), 040.
[24] C. Dunkl. Vector-valued Jack polynomials and wavefunctions on the torus. J.Phys. A: Math. Theor. 50 (2017) 245201 (21 pp).
[25] P. Etingof. Calogero-Moser systems and representation theory. Zurich Lecturesin Advanced Mathematics, 2007.
[26] P. Etingof. Cherednik and Hecke algebras of varieties with a finite group action.(2004), arXiv:math/0406499v3.
[27] P. Etingof. Proof of the Broué-Malle-Rouquier conjecture in characteristic zero(after I. Losev and I. Marin-G. Pfeiffer). arXiv:1606.08456, 2016.
[28] P. Etingof. Supports of irreducible spherical representations of rational Cherednikalgebras. Adv. Math. 229(3) (2012), 2042-2054.
[29] P. Etingof, V. Ginzburg. Symplectic reflection algebras, Calogero-Moser space,and deformed Harish-Chandra homomorphism. Invent. Math. 147 (2002), 243-348.
198
[30] P. Etingof, E. Gorsky, I. Losev. Representations of Cherednik algebras with min-imal support and torus knots. Adv. Math. 227 (2015), 124-180.
[31] P. Etingof, X. Ma. Lecture notes on rational Cherednik algebras. arXiv:1001:0432.
[32] P. Etingof, E. Rains. Central extensions of preprojective algebras, the quantumHeisenberg algebra, and 2-dimensional complex reflection groups. J. Algebra 299(2006), no. 2, 570-588.
[33] P. Etingof, E. Stoica, with an appendix by S. Griffeth. Unitary representationsof rational Cherednik algebras. Represent. Theory 13 (2009), 349-370.
[34] M. Feigin. Generalized Calogero-Moser systems from rational Cherednik algebras.Selecta Math. (N.S.) 18 (2012), N1, 253-281.
[35] M. Geck. A note on Harish-Chandra induction. Manuscripta Math. 80 (1993),393-401.
[36] M. Geck, G. Hiss, F. Luebeck, G. Malle, G. Pfeiffer. CHEVIE - a system for com-puting and processing generic character tables. Applicable Algebra in EngineeringComm. and Computing 7 (1996), 175-210.
[37] M. Geck, G. Pfeiffer. Characters of finite Coxeter groups and Iwahori-Heckealgebras. London Math. Soc. Monographs, New Series 21. Oxford University Press,New York (2000).
[38] V. Ginzburg, N. Guay, E. Opdam, and R. Rouquier. On the category 𝒪 forrational Cherednik algebras. Invent. Math. 154 (2003), 617-651.
[39] I. Gordon, S. Griffeth. Catalan numbers for complex reflection groups. Am. J.Math. 134 (2012), 1491-1502.
[40] I. Gordon, M. Martino. Monodromy of partial KZ functors for rational Cherednikalgebras. Symmetries, Integrable Systems and Representations (2013), 133-154.
[41] I. Gordon, J. T. Stafford. Rational Cherednik algebras and Hilbert schemes. Adv.Math. 198 (2005), no. 1, 224-274.
[42] I. Gordon, J. T. Stafford. Rational Cherednik algebras and Hilbert schemes. II.Representations and sheaves. Duke Math. J. 132 (2006), no. 1, 73-135.
[43] E. Gorsky, A. Oblomkov, J. Rasmussen, V. Shende. Torus knots and the rationalDAHA. Duke Math. J. 163, no. 14 (2014), 2709-2794.
[44] S. Griffeth. Unitary representations of cyclotomic rational Cherednik algebras.Preprint, arXiv:1106.5084.
[45] S. Griffeth, A. Gusenbauer, D. Juteau, M. Lanini. Parabolic degeneration ofrational Cherednik algebras. arXiv:math/1502.08025v1.
199
[46] R. B. Howlett, G. I. Lehrer. Induced cuspidal representations and generalizedHecke rings. Invent. Math. 58 (1980), 37-64.
[47] R. Howlett. Normalizers of Parabolic Subgroups of Reflection Groups. J. LondonMath. Soc. (2) 21 (1980), no. 1, 62-80.
[48] J. E. Humphreys. Reflection groups and Coxeter groups. Cambridge Studies inAdvanced Mathematics, 29, Cambridge University Press (1990).
[49] Y. Ilyashenko, S. Yakovenko. Lectures on Analytic Differential Equations. Grad-uate Studies in Mathematics, vol. 87, Amer. Math. Soc., Providence, RI, 2008.
[50] J. C. Jantzen. Modulen mit einem höchsten Gewich. Lect. Notes in Math. 750,Springer: Berlin-Heidelberg-New York (1979).
[51] I. Losev. Completions of symplectic reflection algebras. Selecta Math. 18 (2012),N1, 179-251.
[52] I. Losev, S. Shelley-Abrahamson. On refined filtration by supports for rationalCherednik categories 𝒪. Selecta Math. (N.S.) 24 (2018), no. 2, 1729-1804.
[53] I. Losev. Finite dimensional quotients of Hecke algebras. Algebra Number Theory9 (2015), 493-502.
[54] I. Losev. On isomorphisms of certain functors for Cherednik algebras. Repres.Theory, 17 (2013), 247-262.
[55] I. Losev. Supports of simple modules in cyclotomic Cherednik categories 𝒪. (2015)arXiv:1509.00527.
[56] G. Lusztig. Characters of reductive groups over a finite field. Ann. of Math. Stud.107, Princeton University Press, 1984.
[57] I. Marin, G. Pfeiffer. The BMR freeness conjecture for the 2-reflection groups.arXiv:1411.4760, to appear in “Mathematics of Computation”.
[58] J. Michel. The development version of the CHEVIE package of GAP3. J. ofAlgebra 435 (2015), 308-336.
[59] M. Murin, S. Shelley-Abrahamson. Parameters for generalized Hecke algebras intype 𝐵. Preprint, arXiv:1708.03673v2.
[60] K. Muraleedaran. Normalizers of parabolic subgroups in unitary reflection groups.PhD thesis, University of Sydney, 2005.
[61] E. Norton. Irreducible representations of rational Cherednik algebras for excep-tional Coxeter groups, part I. (2014) arXiv:1411.7990.
[62] E. M. Opdam. A remark on the irreducible characters and fake degrees of finitereal reflection groups. Invent. Math. 120 (1995), 447-454.
200
[63] R. Rouquier. 𝑞-Schur algebras and complex reflection groups. Mosc. Math. J. 8(2008), 119-158.
[64] H. L. Royden, P. M. Fitzpatrick. Real Analysis, Fourth Edition. Prentice Hall,Boston (2010).
[65] P. Shan. Crystals of Fock spaces and cyclotomic rational double affine Heckealgebras. Ann. Sci. Ecole Norm. Sup. 44 (2011), 147-182.
[66] P. Shan, E. Vasserot. Heisenberg algebras and rational double affine Hecke alge-bras. J. Amer. Math. Soc. 25 (2012), 959-1031.
[67] S. Shelley-Abrahamson. The Dunkl weight function for rational Cherednik alge-bras. Preprint, arXiv:1803.00440.
[68] S. Shelley-Abrahamson, A. Sun. Towards a classification of finite-dimensionalrepresentations of rational Cherednik algebras of type 𝐷. To appear in Journal ofAlgebra and Its Applications (2018)
[69] G. C. Shephard, J. A. Todd. Finite unitary reflection groups. Can. J. Math. 6(1954), 274-304.
[70] V. Venkateswaran. Signatures of representations of Hecke algebras and rationalCherednik algebras. Preprint, arXiv:1409.6663v3.
[71] D. Vogan. Unitarizability of certain series of representations. Ann. of Math. 120(1984), 141-187.
[72] W. Wasow. Asymptotic Expansions for Ordinary Differential Equations. Krieger,Huntington, N.Y. (1976) (orig. Wiley, 1965).
[73] S. Wilcox. Supports of representations of the rational Cherednik algebra of type𝐴. arXiv:math/1012.2585v2.
[74] W. L. Yee. Signatures of invariant Hermitian forms on irreducible highest-weightmodules. Duke Math. J. 142 (2008), no. 1, 165-196.
[75] Maciej Zworski. Semiclassical analysis. Graduate Studies in Mathematics, vol.138, Amer. Math. Soc., Providence, RI, 2012.
201